EvaluatingDriver DistractionCountermeasures

Master Thesis in Cognitive Science

LIU–KOGVET–D– –04/08– –SE

2004–06–16

Author: Rikard Karlsson

Cognitive Science Study Program

Linköpings universitet, Sweden

Supervisor & Examiner: Prof. Erik Hollnagel

Cognitive Systems Ergonomics lab

Department of Computer and Information Science

Linköpings universitet, Sweden

Supervisor: Ph.D. Hans-Erik Pettersson

Swedish National Road and Transportation Institute

Linköping, Sweden

ABSTRACT

iii

AbstractStatistics showing that in-vehicle driver distraction is a major contributing causein road accidents is presented. Driver distraction is defined building on the drivingtheory by Gibson and Crooks. The idea to use driver distraction countermeasuresas a way of mitigating the effects of the driver distraction problem is thenintroduced. A requirement list is formulated with ten requirements thatdistraction countermeasures should meet. A simplification of regarding distractionas a gaze direction problem makes way for designing an experiment to evaluatetwo driver distraction countermeasures in which new eye-tracking technologyplays a key role. The experiment also makes use of a simulator, a surrogate in-vehicle information system as a distractor, and thirty subjects. The mostimportant dependent measures were in-vehicle glance time and a steering wheelreaction time measure. The evaluated countermeasures – a blue flash at middle ofthe road position and a kinesthetic brake pulse – could, however, not be shown tomeet the most important of the requirements formulated. The lack of effect of thecountermeasures in the experiment may either depend on their actual inefficiencyor on methodological shortcomings of the experiment. These alternatives arediscussed. It is speculated that the biggest problems with the possible lack ofactual efficiency have to do with that the theoretical basis for using a flash did nottransfer to the driving setting, and that the brake pulse used was too weak. Themethodological problems have to do with the non-validated dependent measuresused, missing data, nuisance warnings, insufficient distractors, non-precisehypotheses, and difficulties with separating the effect of the countermeasuresfrom the psychological force to look on the road.

Keywords: Driving Behavior, Accident Prevention, Attention, Distraction,Simulators, Eye Movements

PREFACE

v

PrefaceThis master thesis has been written as a part of the studies at the cognitivescience program in Linköping, Sweden. The thesis has also been a part of a jointventure with the Swedish National Road and Transportation Institute (VTI),SAAB Automobile AB and SmartEye AB. The project was funded by SAABAutomobile and Vinnova. The purpose of the project was to contribute withknowledge about and develop driver distraction countermeasure(s) that redirectsthe gaze of a visually distracted driver to the road in order to increase trafficsafety.

My supervisors have been Prof. Erik Hollnagel, at the cognitive systemsergonomics lab at Linköpings universitet and Ph.D. Hans-Erik Pettersson, at VTI.

Thank you, Lena Nilsson (VTI) and Arne Nåbo (SAAB) for letting me participatein this very interesting project. Thank you, Håkan Jansson (VTI) and MikaelAdlers (VTI), for being so helpful with all the programming. Thank you, PerSörner (SmartEye) for giving such good support on the eye-tracking system, andthank you Leif Lannto (VTI) for helping out with the “stimuli hardwarepreparation”. Thank you, Beatrice Söderberg (VTI), and Fredrik Nilsson (VTI) forbeing such wonderful experimenters. Thank you, Elisabeth Ågren, BirgittaThorslund, and Sara Nygårdhs for being there to discuss the hardships of writinga master thesis. Thank you Karin and Jonas for providing good times and goodmeals at your place after long hours in front of the computer. Finally, thank youGunilla for the support you have given me in the last phase of this thesis work.

CONTENTS

vii

ContentsABSTRACT............................................................................................................... III

PREFACE................................................................................................................... V

CONTENTS ............................................................................................................. VII

List of Figures.......................................................................................................... ixList of Tables ........................................................................................................... ix

INTRODUCTION ....................................................................................................... 1

The epidemiology of driver distraction ................................................................... 1The driver distraction concept................................................................................. 4

Driving.................................................................................................................. 4Distraction............................................................................................................ 5Some remarks and a simplification .................................................................... 7

Understanding the distracted driver ...................................................................... 8An individual case................................................................................................ 8The psychological force to look on the road........................................................ 9Contention scheduling in attention .................................................................. 10

The countermeasure method to reduce in-vehicle distraction ............................ 11Previous research............................................................................................... 11Countermeasure requirements ......................................................................... 13Distraction criterion .......................................................................................... 14

Hypotheses.............................................................................................................. 15Blue Flash .......................................................................................................... 15Brake Pulse ........................................................................................................ 16

Methodological issues ............................................................................................ 16Using an experiment ......................................................................................... 16Using a simulator............................................................................................... 17Using eye-tracking equipment .......................................................................... 17Using a surrogate distractor: the S-IVIS (experiment distractor I) ............... 18Using an ecological distractor: the glove box paper fall-out (experimentdistractor II) ....................................................................................................... 20Using an experimental distraction criterion.................................................... 20Using steering reaction and glance duration times as dependent measures 21Using other elicitation techniques .................................................................... 21

METHOD ...................................................................................................................23

Equipment .............................................................................................................. 23The VTI simulator.............................................................................................. 23The S-IVIS task distractor ................................................................................ 23The glove box distractor .................................................................................... 24The Smart Eye system and IDP........................................................................ 24The Flash countermeasure................................................................................ 25The Brake countermeasure ............................................................................... 25

Subjects ................................................................................................................... 26Procedure ................................................................................................................ 26

CONTENTS

viii

Pilot experiment .....................................................................................................28

RESULTS...................................................................................................................29

Glance duration time after condition ....................................................................29Steering reaction time............................................................................................30Glance duration times overall ...............................................................................30

Average glance times after gender and age .....................................................31S-IVIS answers .......................................................................................................31Glove box distractor glance behavior ....................................................................32Questionnaire .........................................................................................................32Video recordings .....................................................................................................33

DISCUSSION ............................................................................................................35

Inefficient countermeasures ..................................................................................36Methodological shortcomings ................................................................................36

Glance duration time .........................................................................................36Steering reaction time .......................................................................................37Visual processing time.......................................................................................38Missing data .......................................................................................................39Nuisance warnings.............................................................................................39Insufficient distractors ......................................................................................39Non-precise hypotheses .....................................................................................40Lack of separation between the effect of the countermeasures and thepsychological force to look on the road..............................................................40

A comprehensive attention support system..........................................................41

CONCLUSION ..........................................................................................................42

REFERENCES..........................................................................................................43

APPENDICES...........................................................................................................47

A – Translated written instructions......................................................................47B – Translated questionnaire with results ...........................................................48C – Diagrams ..........................................................................................................56D – Glance duration times for glove box...............................................................57

CONTENTS

ix

List of FiguresFigure 1: Driver distraction statistics of Stutts et al. (2001)..................................... 3Figure 2: Glance duration histogram of Hada (1994) .............................................. 10Figure 3: Examples of responses to S-IVIS............................................................... 19Figure 4: Examples of difficulty levels of S-IVIS tasks............................................ 20Figure 5: Experiment setup ....................................................................................... 24Figure 6: SmartEye screen with IDP ........................................................................ 25Figure 7: Hypothetical relation between variables .................................................. 27Figure 8: Glance duration results.............................................................................. 29Figure 9: Steering reaction results............................................................................ 30Figure 10: Glance duration result histogram ........................................................... 31Figure 11: Result relation between variables ........................................................... 38Figure 12: Components of an attention support system.......................................... 41

List of TablesTable 1: Distraction criterion of Holmström & Johansson, 2003 ............................ 15Table 2: Questionnaire results................................................................................... 32Table 3: Video results ................................................................................................. 33

INTRODUCTION

1

IntroductionThere has always been a concern for how new technology can lead to drivers thatdo not focus on the road. Around 1915 the concern was about how windshieldwipers might cause drivers to be distracted or hypnotize them by their repetitivemonotonic motion. Later, in the 1930:s when radios were introduced in cars somestates in the US considered to ban them because of their alleged distractibility(GM, 2003). Today we see a proliferation of in-vehicle information systems (IVIS)such as telephones, navigation systems, email, Internet services and eventelevision in cars and an accompanying concern for their distractibility.

Recent statistics material suggests that this concern is justified as will be shownbelow. It is shown that in-vehicle distractors constitute the largest source of driverdistraction. This concern leads car manufacturers to engage in research to comeup with new knowledge about driver distraction and technology that can mitigatethe hazardous effects of new (and old) IVIS. One way to do this is to develop driverdistraction countermeasures. Driver distraction countermeasure ideas arebecoming extra interesting because of the unobtrusive eye-tracking technologythat just recently have entered the market. This thesis is one of the first academicworks that uses unobtrusive eye-tracking technology to evaluate driver distractioncountermeasures and it has therefore been of an explorative nature when it comesto methodology and technology. The aim has, however, been to contribute withknowledge so that future research may be facilitated.

The epidemiology of driver distractionRoad accidents that lead to fatalities and injuries are a global health concern(Peden, 2001). To find out exactly how big the problem is, is not easy, but somesources claim that about 3,000 people are killed and 30,000 are wounded everyday on the world’s roads (Norton, Hyder, & Peden, 2001; Roberts, Mohan, &Abbasi, 2002). The concern for traffic safety is thus justified. In Sweden, where551 people were killed and 22,913 were injured during 2001 (SIKA/SCB, 2002),this concern is manifest in the radical governmental policy of the Vision Zeroprogram (SNRA, 2003), with the goal to reduce fatalities and serious injuries intraffic to nothing. This policy has led to big efforts in infrastructure development,but the policy also puts demands on traffic research to explore every possiblecause of traffic accidents. Here, the focus will be on one of these possible causes;driver distraction, and especially on in-vehicle distraction. Before a presentation ismade of available statistics on driver distraction just a few things should bementioned about accident causation.

To establish the cause of an automobile accident is difficult. It is never only oneparticular event that causes an effect. It is always a complex of conditions thatlead up to an effect, and to complicate it further, this complex can be differentevery time it causes the effect. This is basically what Mackie (1965) means withhis causation theory in which he introduces the notion of the INUS condition (anInsufficient but Necessary part of a condition that in itself is Unnecessary butSufficient for the result). The following fictional yet illustrative short story couldserve as an example: “A young driver is in a hurry to get to his girlfriend’s house,and is driving at 80 mph on a 65 mph road just after a rainstorm has passed. He is

INTRODUCTION

2

also busy operating his newly installed radio. Suddenly, a deer comes up from theleft ditch of the road. When our driver sees the deer he steers the car off the road,hits a tree, and is fatally wounded.” Is it because he is speeding, or because theroad is wet, or because there are trees beside the road, or because he is operatingthe radio, or because the radio has a strange user interface design, or because theradio is located far from the steering wheel, or because the driver is young andinexperienced, or because he is a man, or is it just because he is driving a car, thatthe young man dies in this example? Of course, all these conditions (and manymore) taken together could be said to cause the result of the tragic story. That thedriver was distracted by the radio is an insufficient, but as we look upon it indriver inattention and distraction research, a necessary or at least contributingcondition for the driver to hit the tree. Taken together with other conditions thewhole is sufficient to the result of the accident. There are many other conditionsthat have to be fulfilled for there to be an accident, the distraction by itself is notenough. Nevertheless, to get anywhere in a research attempt to increase trafficsafety we have to delimit the problem and disregard some conditions and look onone recurring condition, which we think is highly relevant. In this thesis it is thecontributing condition that in-vehicle distraction means that will be underconsideration.

It is not easy to find out how many accidents that involve in-vehicle distraction.Indeed, it is generally very difficult to establish the different conditions ofaccidents that have to do with pre-crash driver behavior. If asked about what wasgoing on before an accident, the driver might have forgotten, does not want to tell,or is unable to express her attentional state prior to the accident. Wang, Knipling,& Goodman (1996) put it like this: “Crash investigation is inherently aretrospective, reconstruction process rather than an empirical process. There areno “instant replays.” Therefore, even the best and most in-depth crashinvestigations are, to some extent conjectural.” (pp. 14-15).

A number that is often mentioned in the literature is that 25% of crashes dependson driver distraction/inattentiveness. This figure probably has “its roots” in astudy carried out by Wang et al. (1996) which states that 25.6% of the towawaycrashes of 1995 in the US involved inattentive drivers with 13.2% specificallyinvolving distraction. This figure is based on the same CDS data that Stutts,Reinfurt, Staplin, & Rodgman (2001) used in a more recent study. In this laterstudy, Stutts et al. report that 8.3% of the total number of accidents involving atleast one passenger vehicle that was towed away from the crash scene, had driverscategorized as distracted. This statistic is based on the US National AccidentSampling System (NASS) Crashworthiness Data System (CDS) for the yearsbetween 1995 and 1999 (i.e. not only 1995). Professional field researchers thathave been trained to fill out a form with predefined categories, gather informationfrom a sample of accidents. The sample consists of approximately 5,000 police-reported crashes annually and is weighted to reflect the total number of UStowaway crashes. Most of the information that is collected in each investigationdeals with, as the database name suggests, how the crashed cars physicallymanaged the accident, but since 1995 the form also has a special section devotedto driver inattention and distraction. In this way it may be possible to makeapproximate inferences about the overall incident rate of driver distraction relatedaccidents. To see how Stutts et al. speak about the distraction definition see thenext section on the distraction concept (p. 5).

INTRODUCTION

3

Out of an average total number of 3,420,000 drivers in US towaway crashes peryear during 1995-99, 284,000 (the 8.3%) were distracted according to the CDSprotocol. Of these drivers, 7.9% were seriously or fatally injured which translatesinto approximately 22,400 people. Stutts et al. (2001) further divide the distracteddriver category into different types of distraction sources: “Outside person, objector event” - 29.4%, “Adjusting radio, cassette, CD” - 11.4%, “Other occupant invehicle” - 10.9%, “Moving object in vehicle” - 4.3%, “Other device/object broughtinto vehicle” - 2.9%, “Adjusting vehicle/climate controls” 2.8%, “Eating or drinking”- 1.7%, “Using/dialing cell phone” 1.5%, “Smoking related” 0.9%, “Otherdistraction” - 25.6%, “Unknown distraction” - 8.6%. Note that 37% of all thedistraction can be classified as in-vehicle distraction (see Figure 1).

Figure 1: Driver distraction statistics of Stutts et al. (2001)

There is more research that support the Stutts et al. (2001) result that a largepart of what is classified as distraction related crashes involves in-vehicledistraction. An accident narratives data base was searched using keywords byWierwille & Tijerina (1996) with North Carolina data from 1989 and 1992. Of189,464 narratives 2,816 (1,5% of all narratives) were relevant citations, i.e.narratives that really had to do with distraction. Of these relevant citations, 55%of all the distraction can be classified as in-vehicle distraction. In an Australianstudy with statistics from New South Wales, presented by Lam (2002), it is shownthat all age groups except 40-49 year olds are at greater risk to be involved in aninside-vehicle distraction related crash than they are to be involved in a nodistraction related crash. The risk of being involved in an outside vehicledistraction crash was less than both inside-vehicle distraction crash and nodistraction crash.

All of the presented statistics have clear weaknesses. The NASS/CDS numbersmust be considered in light of the fact that of the total number of estimated UStowaway crashes, 35.9% of the involved vehicles had drivers whose attentionalstate was categorized as unknown. This is a very large number, making the other

05

10152025303540

Outsideperson,

object orevent

In-vehicledistraction

Otherdistraction

Unknowndistraction

Distraction category

Shar

e of

tota

l (%

)

INTRODUCTION

4

numbers unreliable. However there are reasons to believe that the problem is notthat too many accidents are reported as involving distraction, rather the problemseems to be that of underreporting. Of the 48.6% that were reported as attentivein the Stutts et al. study one might suspect that some were maybe not attentiveafter all. Wang et al. (1996) made the interesting observation that in their 1995data, 45.8% of the drivers who were categorized as attentive did not attempt anavoidance maneuver, which, as they put it, “…is somewhat surprising in view ofthe what would be expected from an attentive driver.” (p. 13). For the narrativesstudy, the problem of underreporting is even worse. With open-ended narratives itis even more uncertain that distraction cases are captured. Some distractionsources are perhaps easier to write about in the police reports while others do notget any attention at all from the police officers. More important, however, isprobably road users’ unwillingness to report their own negligence or carelessness,as they might feel a lapse in attention is. Exactly how prevalent driver distractionis as a contributing factor for accidents is difficult to say, nonetheless, it is vital torecognize that driver distraction is an important contributing factor in road trafficaccidents. Indeed, despite all the weaknesses in the statistics material presentedabove, it is clear that different types of unsafe driver distraction do exist, anddifferent studies suggest that in-vehicle distraction seems to constitute asubstantial portion of the total driver distraction problem.

The driver distraction conceptWe have now seen that many accidents on public roads could be partlyattributable to driver distraction, but what does the driver distraction conceptmean? That is the question for this section to answer. The first step will, however,not be to plunge into a discussion of distraction, because first it is vital to have agrasp of what safe driving is about.

DrivingThe primary driving task is a human factors concept that is widely used to denotewhat driving is from a safety point of view. A problem with the concept, however,is that it is often only loosely referred to as something the driver should beoccupied with in contrast to secondary tasks, such as talking on the phone, usingthe navigation system, or tending a baby. Another problem is that to speak of oneprimary task might be misleading since some authors speak about several drivingtasks. Here a complete presentation of all the subtasks in driving will not bepresented, since the objective of this thesis is different. Presentations in athorough style of driving tasks have been made elsewhere (McKnight & Adams,1970; Michon, 1985). The matter of interest here is only to have some form oftheoretical framework that provides a conception of the primary driving task(s) ata proper level of detail. To serve our purposes, this framework should also accountfor the driver distraction concept. The analysis of Gibson & Crooks in their paperfrom 1938 (Gibson & Crooks, 1938) may be such a theoretical framework with auseful level of detail. Somewhat simplified, the goal of safe driving, and hence theprimary driving task, is to always be able to steer or stop so that collisions areavoided. Gibson & Crooks call their study a “Theoretical field-analysis ofautomobile-driving” in which a whole set of concepts to describe driving ispresented. In this set the concept of ‘field of safe travel’ takes a central position.The field of safe travel is in the direction where the car is headed and boundedlaterally by meeting traffic, ditches and pavements, etc. and frontally by traffic in

INTRODUCTION

5

front of the car. The boundaries of this field are thus made up of obstacles, whichthe driver must avoid. In the field analysis there is also another field or zone thatis called the ‘minimum stopping zone’ which is bounded by the distance that isrequired to come to full stop from the current position and speed if maximumbraking is applied. When the field of safe travel shrinks so that it frontallybecomes nearly as small as the minimum stopping zone, the driver becomesuneasy and adapts by reducing speed to increase the safety margin (differencebetween field of safe travel and minimum stopping zone). If the field of safe travelbecomes smaller than the minimum stopping zone, the driver is in an emergencysituation. In the terms of Gibson & Crooks, the task of the driver is thus to keepthe ratio between the field of safe travel and the minimum stopping zone, thefield-zone ratio, of a proper magnitude. Their hypothesis is that this is mainly aperceptual task and that the steering task “is a perceptually governed series ofreactions by the driver of such sort as to keep the car headed into the middle of thefield of safe travel” (p. 456). Braking and accelerating are also in the same manner‘governed’ by perception. It is by perceiving and attending to the road that thetasks of driving may be properly accomplished. In the terms of Gibson and Crooksthere is actually one predominant driving task and that is to maintain a visualfield1 so that the field of safe driving and the minimum stopping zone are correctlybounded. This is also how the primary driving task is defined in this thesis.

DistractionWhen the driver’s visual field is not properly maintained, because of the driver’svisual engagement into non-driving related events, a mismatch between thebehavioral field of safe travel and the objective field of safe travel might be theresult. Gibson & Crooks (ibid.) phrase it like this:

Inattentiveness in the driver usually means thatobjects in the terrain or inside the car which arenot pertinent to locomotion stand out in his visualfield and that consequently his field of safe travel,if it exists at all, may become incorrectly bounded.(p. 458)

This definition of inattentiveness is very close to what is meant by distraction inthis thesis. It should only be complemented with two details to work as this thesis’distraction definition. The first is that there is a time aspect involved indistraction; it is only after some time has passed of engagement into things notpertinent to safe driving that the field of safe travel may become incorrectlybounded. The second is that the non-pertinent objects must be external to thedriver. These two details are emphasized in a much more recent driver distractiondefinition provided by Stutts et al. (2001):

AAAFTS2 has chosen to focus its efforts specificallyon driver distraction, rather than the broadercategory of driver inattention. It defines distractionas “when a driver is delayed in the recognition of

1 Gibson & Crooks (ibid.) use the visual field concept to denote the awareness of the driving situationthat the driver has acquired through perception and attention. We will later call this ‘recognition ofinformation’.2 American Automotive Association Foundation for Traffic Safety

INTRODUCTION

6

information needed to safely accomplish thedriving task because some event, activity, object, orperson within or outside the vehicle compelled ortended to induce the driver’s shifting attentionaway from the driving task”. The presence of atriggering event distinguishes a distracted driverfrom one who is simply inattentive or “lost inthought”. (p. 6)

Here, a triggering event, activity, object, or person in the driver’s environment willalso be called a distractor3.

The first thing we notice with this distraction definition is that, a “delay in therecognition of information” is needed for a person to be considered as distracted.That is, the time during which the visual field of the driver is not updated, inorder to correctly bound the field of safe travel and the minimum stopping zone, iswhat turns a driver into a distracted driver. If it is enough for this delay to be onlyslightly above 0 s for a driver to be distracted or whether the delay must be longeris something that is not discussed by Stutts et al. (ibid.). We will, however, returnto this issue in the discussion of a distraction criterion below (p. 14). It is alsoworth notice that to define distraction along the lines of recognition delay alsomeans that a total failure of the primary driving task, e.g. driving off the road, isnot needed for a person to be called distracted.

The second observation we may make of this definition is that the distractor mustbe external to the driver. An inattention source from ‘within’ the driver is not adistractor which means that a driver who is simply inattentive or ‘lost in thought’is not distracted. Neither are states like drowsiness, preoccupation with competingthoughts, daydreaming, worrying, being caught in affection, etc. regarded as casesof distraction. This is contrary to many other definitions of attention anddistraction. William James states, e.g. that “[attention] is a condition which has areal opposite in the confused, dazed, scatterbrained state which in French is calleddistraction, and Zerstreuheit in German.” (James, 1890/1981, p. 382). The“confused, scatterbrained state” is something that Stutts et al. would callinattention and not distraction. Using this definition of driver distraction meansthat there must always be a distractor in the driver’s environment that causes ashift of attention away from the primary driving task. That the attention isdirected away from the primary driving task is invariably the effect of thedistractor. Another way to say this is that the driver will become distracted if andonly if there is a distractor present in the driver’s environment.

The inattentiveness definition of Gibson & Crooks and the driver distractiondefinition of Stutts et al. taken together show to be a fruitful combination. TheGibson & Crooks definition relates our issue to a comprehensive driving theoryand the Stutts et al. definition introduces the time aspect and makes a distinctionbetween inattention and distraction by saying that distraction is the result oftriggering events external to the driver. The distraction definition of this thesis istherefore: Distraction occurs when a driver is delayed in the recognition ofinformation needed to accurately bound the field of safe travel and the minimum

3 Distractor rather than distracter since the latter is more a person and the former could be anything, ageneral event.

INTRODUCTION

7

stopping zone because an event (a distractor) external to the driver compelled ortended to induce the driver’s shifting attention away from this information.

The discussion up to this point helps in defining what a driver distractioncountermeasure is. A driver distraction countermeasure redirects the attention of adistracted driver from the distractor to information that is necessary to bound thefield of safe travel and the minimum stopping zone.

Some remarks and a simplification

DRIVING AND MOTOR FUNCTIONS

According to our definition, distraction affects driving in a negative mannerprimarily at a perceptual level (“recognition of information”). The definitionimplies that motor functions are not directly impaired by distraction. However,impairment of perception usually leads to improper motor responses and theseimproper motor responses are, indeed, the unwanted unsafe behavior that leads toaccidents. Nevertheless, here we have the belief that these motor responses areperceptually governed, and therefore it is the perceptual aspect of the problemthat should be in focus.

THE CONTEXTUAL DEPENDENCY OF DISTRACTION

Whether an event is categorized as a distractor is dependent on the driver and thesituation. Something that is distracting to one person at one time is perhaps notdistracting to another person (or the same person) at another time. We will returnto this discussion when we speak about experiment distractors (p. 18).

A SIMPLIFICATION: ATTENTION AS GAZE DIRECTION AND A DISTINCTION TO LOOK BUTDID NOT SEE

In our distraction definition the attention shift away from the pertinentinformation is central. Changing gaze direction is often linked to such anattentional shift. Theoretically, it is however possible to direct attention todifferent parts of the field of view. However, since the fovea is only 2 degrees ofvisual angle, gaze is for much of the time driven by our need to attend (Wickens &Hollands, 2000). That is, to attend to different things while driving, gaze has to beredirected most of the time.

Sometimes it is, however, not enough to gaze on something in order to payattention to it. The “looked but did not see”-phenomenon means that the driverdirects the gaze on something without it being included, as it should have been, inthe visual field of driving. When an obstacle or a hindrance in front of a driver’sown car is run into, although the driver is looking at it, then we have an exampleof “looking without seeing”. The “looked but did not see”-phenomenon can forinstance be studied in change blindness experiments, (see e.g. Rensink, 2000) andis also one of the inattention categories of the NASS/CDS statistics (see p. 3).However, the driver distraction concept does not involve the “looked but did notsee”-concept.

In other words, there are two relevant aspects of a driver’s attention. The firstaspect is that the driver must direct the gaze on pertinent information forlocomotion. If the driver does not, s/he is distracted (by an external object) orinattentive (by an internal state). The second aspect of the attention problem of

INTRODUCTION

8

driving is that it is not enough to direct the gaze; the driver might despite the facts/he is looking at the road not be attending to pertinent traffic events, and thens/he is looking without seeing. Thus, whereas the “looked but did not see”-phenomenon might be seen as a cognitive attention problem, driver distractionmight be seen as a gaze direction problem. Since the focus of this thesis is ondriver distraction, it is gaze direction that will be studied here, whereas thecognitive aspects of attention that the “look but did not see”-phenomenonhighlights will be disregarded for the time being.

With the simplification of equating attention with gaze direction and disregardingthe “looked but did not see”-phenomenon we may rephrase our previousdefinitions:

Distraction occurs when a driver is delayed in the recognition of informationneeded to accurately bound the field of safe travel and the minimum stopping zonebecause an event (a distractor) external to the driver compelled or tended to inducethe driver’s shifting gaze away from this information.

A driver distraction countermeasure redirects the gaze of a distracted driver fromthe distractor to information that is necessary to bound the field of safe travel andthe minimum stopping zone.

Understanding the distracted driverBehind the numbers that were presented in the statistics section above (p. 1) aremany individual cases where one or more drivers were distracted. To betterunderstand the distraction problem, this section will first present an individualcase of driver distraction taken from real life that led to an accident. Then thepsychological force that does not suffice when drivers are involved in distractionaccidents is discussed. This section ends with a presentation of a theory that mayhelp to explain what happens when a driver is distracted.

An individual caseA fictitious story was told above (p. 1) about a young driver on the way to hisgirlfriend’s house – this is one example of an individual case, but it would be muchmore interesting to have accounts of real accidents. However, it is often consideredas bad ethical practice to publicize real-life accident reports. The SwedishNational Road Administration regards their accident reports as highly sensitivepersonal information and do not allow their reports to be made public so thatindividuals’ identity could be revealed (SNRA, 2002). The American NationalTransportation Safety Board does, on the other hand, make accident reportsavailable on the WWW. Here follows an excerpt from an accident report involvinga truck-driver, a clipboard, and a school bus.

Injuries

2 serious, 33 minor injuries

Accident Description

The truckdriver stated that he was traveling easton U.S. Highway 290 about 56 mph when he strucka stopped school bus on Nicholson Lake Road. (Seefigure 1.) The truckdriver said that prior to the

INTRODUCTION

9

collision he braked and swerved to the left, but theright front of the truck struck the left rear of theschool bus. Two students sustained seriousinjuries; 31 students, the busdriver, and thetruckdriver sustained minor injuries. Both vehicleswere destroyed.

The truckdriver stated that he did not see theschool bus because he was late and had beenlooking on his clipboard for a telephone number tocall a business. He found the telephone numberand was placing his clipboard between the seatswhen he looked up and saw that the school bus wasstopped. He had been awake for 14 1/2 hours and,except for a 45-minute nap, had been on duty for 131/4 hours and had been driving for almost 10hours. By the time the driver would havecompleted his shift, he would have been awake for18 hours, on duty for 15 hours, and driving for 12hours. When asked how he felt, the driver statedthat he was "physically tired" but believed that hewas mentally alert.

Probable Cause

The National Transportation Safety Boarddetermines that the probable cause of the accidentwas the truckdriver's inattention to driving.Contributing to the accident may have been thetruckdriver's fatigue.

(NTSB, 2002)

In this accident we see that it is perhaps not only distraction that causes theaccident. The driver, although stating otherwise, might have been drowsy as well,but distraction is one of the contributing factors. In this accident it was attendingan in-vehicle clipboard that made the driver look away for too long for thesituation at hand.

The psychological force to look on the roadDespite the story just told it is most natural for drivers to attend to the road,otherwise we would have had many, many more accidents4. To be sure, if deprivedof the visual field the driver soon becomes very uncomfortable. In fact there seemsto be a psychological force not allowing for long glance times off the road. In areview by Green (1999), results from an experiment (Hada, 1994) that hadAmerican subjects that drove on a real road and were asked to look away from theroad as long as they felt safe to do so, are reported. Below are the results from thatexperiment.

4 Automobile driving would probably not have been a human activity at all.

INTRODUCTION

10

Figure 2: Glance duration histogram of Hada (1994)

The median glance times inside the vehicle were 0.86 s on the expressway, 0.81 son the rural road, 0.68 s on the suburban street, 0.79 s on the urban street, and0.87 s while stopped at urban intersections. Less then 5 % of the totally 10,660glances were longer than 2.2 s (95th percentile). The 97.5th percentile was 2.5 s andthe 99th percentile was 3.6 s. Another similar study (Kimura, Osumi, & Nagai,1990) with Japanese subjects that got the instruction to look at in-vehicle targets“for as long as possible until you feel uncomfortable”, reported average glancetimes that were much higher with medians up to 2.0 s. In another review, thistime by Wierwille (1993) that focuses on natural glance behavior (with noinstructions) for in-car controls and displays, results show that single glance timesfall into a relatively narrow range and it is the number of glances that vary within-car task difficulty. The conclusion that Wierwille draws is that drivers usuallydo not, on the average, look into the car for longer than 1.6 s at the time.

The psychological force to look on the road has also been empirically investigated,albeit a little differently, by Senders et al. (1967). These authors argue for apsychological speed limit that is related to the amount of visual information adriver is allowed to receive. In their study it was shown that intermittentlyoccluding a driver’s line of vision so that the driver cannot see the road, makes thedriver slow down.

These studies suggest that drivers have a “psychological force” to look at the road.The reason for this has probably to do with that people are instinctively concernedabout their safety. In fact, it could be argued that it is surprising how reliable thisinstinct is. Nevertheless, all too many accidents occur and the important questionarises “What cancels the psychological force to look on the road?”. A tentativeanswer to that will be discussed now.

Contention scheduling in attentionThere was something with the clipboard that made the truck driver look down, somuch that the primary driving task was not properly carried out. Thepsychological force that urges for look out was in this case too weak.

The action theory by Norman and Shallice (Norman & Shallice, 1986) is useful toexplain conflicts between tasks. According to their theory, people receive sensoryinformation that goes via sensory-perceptual structures to a trigger database that

INTRODUCTION

11

has triggers, which triggers schemas that prompt for certain actions. Differentschemas can be triggered simultaneously and if they do not compete for actionresources, the actions associated with the schemas can be carried outsimultaneously. If the schemas do compete for action resources the schema withhighest activation value will take over through activation and inhibition incontention scheduling. This process may go on with no conscious or attentionalinvolvement, but through the supervisory attentional system (SAS), will-powerand attention can be exerted for activation and inhibition of schemas also. A thirdfactor, which influences schema activation, is motivation that operates parallelwith SAS. This theory elegantly accounts for e.g. the cognitive distinction betweenautomatic and controlled processing, the Stroop phenomenon, capture errors, andeven neuropsychological evidence.

We may formulate the driver distraction problem in terms of the Norman andShallice theory. Schemas that are likely to possess higher activation levels thanthe primary driving task schema are distracting schemas if their gaze actioncomponent is put in conflict with the gaze action component of the primary drivingtask schema. An example of such a conflicting schema is the clipboard operationschema whose gaze action component corresponds to gaze inside the vehicle and isthus in conflict with the gaze action component of the primary driving task. In thecase of our truck driver, not enough activation was received by the “primarydriving task”-schema so therefore the action component of “reading the clipboard”-schema was carried out, much to the dismay of everyone involved at the scene ofthe crash.

The countermeasure method to reduce in-vehicledistractionThere are, at least, two possible main paths to reduce the problem of distraction.The first is to reduce the possible hazardous sources of it. The second is to activelytry to counteract these sources by other stimuli presented to the driver. The firstpossible path, which we may call “workload design”, has already been probed bythe car manufacturing industry and is manifest in the policies of design of cockpit,IVIS user interfaces, and task management systems and also in some states’ andcountries’ law texts. The second possible path, which we may call “attentionsupport” (Victor, 2000), is a comparatively new approach, and is the path that wewill follow in this thesis. The idea here is that if driver distraction occurs, in spiteof attempts to “build it away”, it would be good to have something that realerts thedriver to the road. The advantage with an attention support system in this sense,is that it (at least conceptually) counteracts all possible distraction sources, alsothose that are not pertaining to IVIS:s, e.g. objects brought into the vehicle suchas maps, portable computers, crying babies, food etc. The special form of attentionsupport system that we will study in this thesis is the distraction countermeasuresystem. Here, a distraction countermeasure system is a system that has a way ofmonitoring the driver to make inferences about the driver’s attentional status and ifa distraction criterion is met it activates some distraction countermeasure.

Previous researchIn previous research, distraction countermeasures per se have not been sothoroughly studied. Research on collision avoidance systems has been more

INTRODUCTION

12

common. Here, the difference between collision avoidance systems and driverdistraction countermeasure systems will be considered as a difference in sensortechnology. Collision avoidance systems use vehicle status/environment sensorsthat estimate headway and time to collision (TTC), whereas distractioncountermeasure systems use driver status sensors for instance those that followeye gaze behavior. Although the sensor technology is different, inspiration for thedevelopment of realerting stimuli for a distraction countermeasure system can bedrawn from the collision avoidance literature. Two examples of collision avoidancesystem studies will be presented here, one of which partly is a review. These havebeen chosen since they both involve distracted drivers.

LEE, MCGEHEE, BROWN, & REYES (2002)In this study different timings of warning set-in were studied in a drivingsimulator with both distracted and undistracted drivers. The drivers were exposedto imminent rear-end collisions while being distracted and, not surprisingly, itwas found that early warnings had better effect, in terms of avoided collisions anddecreased impact severity, than had late warnings. The warning that was usedwas a combined auditory and visual signal. The auditory tone was made up of foursound bursts and each of the bursts had four pulses. The bursts came every 110ms and the pulses were separated by 10 ms. The visual display was mounted justabove the instrument cluster and depicted a vehicle colliding with the rear ofanother vehicle.

SHUTKO (1999)In the literature review of Shutko’s master thesis, studies of warning signals ofcollision avoidance systems are presented. A reference to Dingus (1996) is made inwhich a “soft braking” warning was used. This kind of warning has the advantagethat it warns the driver at the same time as the vehicle slows down a little.Furthermore, the operation of “soft braking” has the advantage of being discrete inthe sense that passengers of the car cannot know whether it is a warning that setsin or if it is the driver who applies the brakes. In the Dingus study the brake forcethat was used was 0.2 g and this helped drivers to avoid collisions.

Other sources presented by Shutko come with recommendations about brakepulses. The first of these sources states that a brake pulse should not startle thedriver or in other ways disrupt driver performance. Moreover, each brake pulseshould be approximately 300 ms. How many pulses or how close in time theyshould come for each warning set-in seems to be something that is not detailed bythis author (Landau, 1995). The second of these sources points out that a brakepulse induces less mental load on the driver, is discrete, and may reduce driverreaction time (Onken, 1994). Finally, there are sources that claim that a brakepulse should be used for forward emergency situations (Weber, Mullins,Schumacher, & Wright, 1994), or as an imminent warning (Wilson, Butler,McGehee, & Dingus, 1996).

Shutko himself used auditory and kinesthetic5 (brake pulse) warnings in aninstrumented commercial truck. The auditory signal was an auditory icon (earcon)that sounded like a tire skid and with speakers located in front of the driver thismight be perceived as a vehicle in front that decelerates suddenly and quickly. 5 Shutko calls the brake pulse a haptic warning. Here, the term kinesthetic is used instead.

INTRODUCTION

13

The kinesthetic signal was implemented so that the brakes were pumped for onesecond with a deceleration of 0.3 g. It was found that the auditory warningdecreased movement times of the foot from accelerator to brake pedal while thebrake pulse decreased the number of collisions and impact speed.

Countermeasure requirementsWe have previously defined what a distraction countermeasure is (p.8); namelythat it is a stimulus that redirects the gaze of a distracted driver to the road.There are, however, other surrounding requirements that apply to a distractioncountermeasure. During the process of writing this thesis a tentative list ofrequirements that apply to driver distraction countermeasures has been developedthat deals with this problem. Such a list works also as an important guideline forthe design of experiments that evaluate countermeasures. The most importantaspect that deals with the definition of a driver distraction countermeasure isfound at the top of the list.

1. Direct gaze to the road. The first requirement is that a countermeasureshould get the driver’s eye gaze back to the road.

2. Enhance ability to cope with critical situation. The driver should behelped by the countermeasures to e.g. avoid collisions or driving off the road.

3. Fast. It should not take long for the countermeasure to have effect aftercountermeasure onset.

4. Unambiguous. The countermeasure stimulus should not be confusing orinterpreted by the driver to mean something that it does not mean.

5. Not distract. It should not delay the driver further in the recognition ofinformation to bound field and zone (see driver distraction definition p. 5) –only direct the driver’s gaze to the road.

6. Not occlude. The physical realization of the countermeasure should notocclude the visual field of the driver.

7. Not auditive. Since it is possible to drive (in Sweden) without hearing, thecountermeasure should not rely on hearing.

8. Accepted. For the countermeasure to be accepted by the driver it should notbe experienced as distressing, upsetting or annoying.

9. Reliable. A countermeasure should work every time and despite largevariations in the ambient environment.

10. Realizable. It should be possible to introduce in mass production in cars at areasonable cost.

The first of these requirements is quite straight forward and follows the definitionwe have of driver distraction that states that distraction is about gaze direction.The second requirement is an important complement to the first, since itemphasizes the reason why drivers should have their gaze on the road. The thirdand fourth requirements relate to the desire that the countermeasure should beintuitive and not require high order cognitive processing resources. Exactly howfast the countermeasure should have effect is not detailed here.

The fifth and sixth requirements point out that there is a possibility that, if notproperly designed, countermeasures may do more harm than good. The seventhrequirement in the list could be discussed since it has been shown that auditorystimuli have effect (Lee et al., 2002; Shutko, 1999) and are usually faster thanvisual stimuli (Wickens & Hollands, 2000) and in most other research andwarning discussions it is taken for granted that the auditory channel might be

INTRODUCTION

14

used to mediate a warning. In this thesis it is, however, noted that the hearingimpaired also may drive so therefore it must be of interest to recognize this factand come up with countermeasures that also this part of the population maybenefit from. The eighth requirement is important since drivers must be willing touse the system, otherwise it is not of much use. The ninth requirement has to dowith that whenever and wherever the driver is categorized as distracted, evenafter many years of use, or with many repeated activations of it, or during anyseason, or time of day in city or on highway, the countermeasures should still haveeffect. The last requirement is of a very practical nature, but not of lessimportance. It is not meaningful to develop countermeasures that cannot beimplemented in cars in real life.

Distraction criterionIn the case of collision avoidance systems (CAS) the criterion for warning set in iscalculated by using sensor information that estimates distance to the vehicle infront and relative speed of the vehicles. In the distraction countermeasure systemof this thesis the sensor information is eye gaze direction and duration. Here adiscussion follows of a simple distraction criterion model that only makes use ofsuch eye gaze information. In this discussion, “information needed to accuratelybound the field of safe travel and the minimum stopping zone” as our driving taskdefinition and distraction definitions formulate it (see p. 5) has been replaced by“ROAD” for reading comfort. A simple, basic idea is that the longer a driver looksaway from the ROAD the more critical it becomes. A distraction criterion for adriver distraction countermeasure system specifies how large this delay can bebefore a driver is categorized as distracted and thus when a distractioncountermeasure should set in.

One could say that the driver has an attentional budget that runs out if the driverlooks away from the ROAD too long, and conversely, when the driver looks on theROAD again the attentional budget receives “funding”. In this model a distractioncriterion is set by three parameters. First, by the initial budget limit when nothinghas been spent on off-ROAD glance time. Second, by the rate at which the budgetruns out when the driver looks away from the ROAD. Third, by the rate at whichthe driver receives funding when looking up again. If the budget runs out, awarning could possibly set in to counteract the budget deficit. Some would saythat the currency of such an attentional budget is information bits (Senders et al.,1967). In a simplified model we could, however, use time units.

To exemplify how it might work, let us assume that the budget is three seconds,i.e. when the driver has looked away from the ROAD for three seconds the systemshould warn. Let us say that a driver wants to find a particular CD in a pile ofCDs on the passenger seat. While driving, s/he looks down on the passenger seatfor 2 seconds (Budget: 1s) but does not find it instantly so s/he looks on the ROADagain, but only for a short time, 0.5 s, too short to notice the pedestrian furtherdown the road (Budget: 1.5s). Then s/he looks on the passenger seat again formore than 1.5s and the alarm sets in just in time for our driver to notice thepedestrian, who has staggered out on the road (and thus has become part of theROAD). Our driver just has time to steer around the pedestrian. Notice that thisdistraction event is divided into two glances away from the ROAD. Note also that inthis example the increment and decrement constants are both set to 1. It is mostprobable that they should be set to something else. One might speculate that theincrement constant should be higher than the decrement constant, i.e. when the

INTRODUCTION

15

driver looks on the ROAD the number of seconds should multiply with a constantthat is more than 1. For instance the increment constant could be 2, then whenthe driver looks on ROAD for 0.7s the budget is increased with 0.7x2 = 1.4s.

There is one study that has used a variant of this distraction criterion model. Inan on-the-road study (Holmström & Johansson, 2003) ten subjects drove for tenminutes in city traffic and ten minutes on suburban road each, with no othertasks. They used the SmartEye eye-tracking equipment (see p. 17) and divided thevisual environment of the driver into six zones: primary, windscreen, maininstrument unit, left rear mirror, right rear mirror, and inner rear mirror. Thedifferent zones had different increment/decrement constants and the constantsalso depended upon road type. Note that this division implicitly relies on theassumption that chances of recognizing “information needed to bound the field ofsafe travel and the minimum stopping zone”6 are highest in the so called primaryzone, which was at approximately middle of the road position.

City SuburbanZone Decrement constant Time (s) Decrement constant Time (s)

Windshield -1.7 2.21 -1.0 3.75

Left rear mirror -1.5 2.50 -1.0 3.75

Right rear mirror -2.0 1.88 -2.0 1.88

Inner rear mirror -1.7 2.21 -1.5 2.50

MIU7 -1.3 2.88 -1.0 3.75

Background -1.7 2.21 -1.0 3.75

Face obstructed -1.0 3.75 -1.0 3.75

Primary zone No limit

Table 1: Distraction criterion of Holmström & Johansson, 2003

The “Background zone” is the zone not covered by the other zones. “Faceobstructed” refers to that the eye-tracking sometimes was obstructed. With thesesettings the ten drivers were categorized as distracted 9 times during a total timeof 10x20 min=3h 20 min. The settings which constitute a driver distractioncriterion were not theoretically supported other than that they built on thecommon sense assumption that the more a driver looks away from the middle ofthe road, measured both in time and angle of eccentricity, the worse it gets.

Hypotheses

Blue FlashAttractive stimuli that might capture a person’s attention have been studied inexperiments (Remington, Johnston, & Yantis, 1992; Yantis & Hillstrom, 1994).

6 Following the definitions that were presented on p. 5.7 Main Instrument Unit (speedometer etc.)

INTRODUCTION

16

From these experiments it is concluded that abrupt visual onsets attract people’sattention, even involuntarily. This makes way for the idea to use abrupt visualonsets to attract a distracted driver’s gaze to the road. Such a stimulus would bedifferent from an ordinary warning in the sense that it would be attractive in itselfand therefore not require any type of interpretation by the driver. An abrupt onsetthat directs the attention of the driver to the road if the driver has been distractedby something should be located at a middle of the road position. This builds on theassumption that chances of finding information needed to accurately bound thefield of safe travel and the minimum stopping zone are highest there (followingour driver distraction definition on p. 7). Even if the driver looks away rathermuch, that is if gaze eccentricity from the middle-of-the road-position is high, itmay be possible that the driver would be attentioned to it. There is muchempirical evidence that peripheral vision is good, perhaps even better than fovealvision, for detecting new perceptual events in the visual field (Dember, 1960). Touse a flashing light mounted on the dashboard so that it is reflected in the windowapproximately at middle of the road position position could be one possiblecountermeasure. Since there are sources (e.g. Graham, 1965) claiming that, in theperiphery, cone sensitivity to short wavelengths is relatively higher than it is inthe fovea, the flash should be blue. A blue flash fulfils requirements 6 (notocclude), 7 (not auditive), and 10 (realizable) of our requirement list (p. 13). Itrequires empirical investigation to test the other requirements, however.

Brake PulseSince there has been previous research showing the efficiency of brake pulses inthe context of collision warning systems, this may also be effective as a driverdistraction countermeasure. A brake pulse fulfils requirements 6 (not occlude), 7(not auditive), and 10 (realizable). Because of its discreetness there are reasons tobelieve that it also will fulfil requirement 8 (accepted). It requires empiricalinvestigation to test the other requirements, however.

Our hypotheses are thus:

H1: A flashing blue light at middle of the road position position is an effectivedriver distraction countermeasure.

H2: A brake pulse is an effective driver distraction countermeasure.

Methodological issuesBefore we go on to the method chapter of this thesis, the different components ofthe methodology used will be discussed. This is done to provide a background forthe methodological choices of the study. All the way, the goal was to make suchchoices so that the different aspects put forth in the requirement list on p. 13should be covered and the hypotheses thus be addressed. The exception is thenineth requirement (reliable) that requires a temporally more extended studythan is possible here.

Using an experimentTo evaluate the efficacy of the proposed distraction countermeasures and test ourhypotheses, the best thing would be to assign people, by random sampling, to twogroups. One group gets the driver distraction countermeasure system installed in

INTRODUCTION

17

their cars and the other (control) group does not get any installation. After a year,or so, the accident statistics of these two groups could be compared and if there isa significant difference that is large enough, according to some predeterminedcriterion, to the advantage of the treatment group, then it could be recommendedto the public to use that countermeasure.

Unfortunately, there are practical problems with this strategy, which means thatthe needed investment for a large-scale field study would not be defendable. First,the two groups must be very large for there to be any measurable differencebetween them, considering the relative small number of incidents that do occur.Second, we are far from knowing what an effective countermeasure system wouldlook like. Third, since there is so little knowledge about driver distractioncountermeasure effects, the safety risks are not known either, and to put driversat risk would not be in line with fundamental ethical research principles. Theresearch process has not come that far, and a large field study lies quite longahead down the road. It is not until more knowledge exists about distractioncountermeasures that such a study would be advisable. At present, much has to bediscovered. Therefore, instead of using a field experiment that would have largerexternal validity, with direct measures such as accident rate, an experiment withindirect measures will be used.

The idea is to use an experimental setting in which subjects drive in a drivingsimulator. While driving in the simulator they are distracted by an in-vehicledistractor that makes the subjects shift their gaze to a head down position insidethe car. When an eye gaze monitor records eye gaze into the car that exceeds adistraction criterion, a distraction countermeasure that should has the effect toredirect the subjects’ gaze to the road is activated. By comparing glance timesfrom the eye gaze monitor and reactions to a critical situation that occurs duringlook down, between conditions where countermeasures are used to a control wherenone are used. Glance time and critical situation reaction are indirect safetymeasures from which inferences about the efficiency of the countermeasures maybe drawn.

Using a simulatorOne of the problems with a large-scale field study is that it is impossible toguarantee the safety of the participants when the effects of the countermeasuresare not known. By conducting experiments in a simulator no such risks exist.Further, conditions can be controlled to a larger degree in a simulatedenvironment compared to real life. A simulator experiment will therefore be themethod by which data is acquired in this study. The problem that simulatorexperiments face has to do with external validity, but at this stage in the researchprocess this problem is considered to be smaller than the economic and safetyproblems of a field study.

Using eye-tracking equipmentA big advantage in this project is that a newly developed eye-tracking system, bySmartEye, was put to disposal. This is a system that uses two cameras that aremounted on the dashboard in the car, IR-lights, and digital image processing, toyield gaze directions of the driver. It is designed to provide information aboutwhere people look. The system looks for facial features and tries to figure out headposition and eye gaze on basis of where these features are located. Car

INTRODUCTION

18

manufacturers are interested in this technical solution since it is unobtrusive, i.e.no eye gaze helmet (see e.g. Harbluk, Noy, & Eizenman, 2000) or anything elsethat the driver has to wear is used. This means that it would work to have such asystem, with cameras, installed as extra equipment in series produced cars. Thecameras would then be used for monitoring the driver for attention supportpurposes, but could also be used for other purposes such as detecting the driver’sattitude and adapt air-bag blow-up, or automatically customize seat adjustment.This eye-tracking equipment is thus not only an interesting research tool, but alsoa commercially viable solution for the car industry.

There has been one previous study (Holmström & Johansson, 2003) that used theSmartEye system for tracking drivers’ eye gaze. The objective of that study wasinitially to determine parameters of a driver attention model by using theSmartEye eye-tracking system, but it was concluded that the SmartEye systemwas in a developing stage and was not reliable enough to be used as a researchtool. Instead, the study focused on evaluation of the SmartEye system. With therecommendations of that study some of the problems have been solved, but therehas not been a thorough validation study in a driving setting, as yet, of theSmartEye system. Together with the SmartEye eye-tracking software, a programhas been developed that can be used for dividing the visual environment of thedriver into different zones. This program was intended to be used to classify theattentional state of a driver and is called Inattention Decision Program, or IDP forshort. It can be used to define zones such as the primary zone (activated whenlooking straight ahead), the center rear mirror zone (activated when looking intothe center rear mirror), the main instrument unit zone (looking at thespeedometer), etc.

Using a surrogate distractor: the S-IVIS (experiment distractor I)A very important aspect of evaluating distraction countermeasures in anexperimental setting is to get experimental control over the distraction itself.Since statistics suggest that in-vehicle distractors constitute the largest portion ofthe distraction problem (see Figure 1, p. 3) we will only focus on those in theexperiment. There are a lot of potential distractors that a driver might besubjected to in a modern car. Crying children, learning how to operate a newradio, operating the navigation system, operating the audio system, picking up anobject from the floor, dialing a phone number on the cellular phone, reading apaper map, etc are all examples of potential distractors. However, differentdistractors have different effect on different people. Something that is verydistracting to one person could be immaterial to another. The problem is furthercomplicated by the fact that the same distractor might affect the same personsdifferently on different occasions. For instance, a newly installed radio may bemuch more distracting to operate than a radio that is well known and understoodby the driver. That is, all of the aforementioned distractors have a greatintrapersonal and temporal variance as to their distractiveness. This poses aproblem in an experimental setting since there is a need for a distractor thatrepeatedly and reliably distracts different people. Moreover, the distractor has tobe calling for continuous attention since what we want to do is to see if ourcountermeasures have effect after a longer period of glance away from the road.That is, its tasks must not be chunkable so that the subject can choose to dividethe task solving in visual chunks.

INTRODUCTION

19

Different suggestions that could be used as a distractor were considered, e.g.visually demanding computer games. The final solution was to use a slightlymodified version of the surrogate in-vehicle information system (S-IVIS) taskdeveloped by Natasha Merat at the Institute for Transport Studies at theUniversity of Leeds, UK8 (Merat, 2003a). The task is designed to simulate thevisual, motor, and cognitive demands on the driver that are comparable to realIVIS (e.g. a navigation system). It also fulfils the requirements of being stablebetween subjects (Merat, 2003b) and demands continuous visual attention.

The subject is prompted by a sound to look at a screen inside the car for every taskpresentation. The task is then to press a yes button when there is an arrow thatpoints upwards in a matrix of arrows, otherwise the no button should be pressed.

Figure 3: Examples of responses to S-IVIS

The original S-IVIS was complemented with an auditive feedback signal whichindicated right (pling) or wrong (beep) answers. This introduced a certain extent ofgame mechanism into the experiment situation and subjects seemed moremotivated to correctly solve the task when auditive feedback was given. To usesuch a motivational cue is theoretically motivated by the Norman & Shallicemodel, described on p. 10 above, since it is not only sensory stimuli and theSupervisory Attentional System that make the subject attend the S-IVIS screenthen, but also motivation. According to the theory, anything that motivates aperson to engage in a schema (here the S-IVIS task solving schema) will increaseits activation value. In this case the heightened activation value of the S-IVISschema results in that the S-IVIS task becomes more distracting9.

The difficulty level of the S-IVIS task is altered in two ways. First, by using threedifferent sizes of the matrix: 4x4, 5x5 and 6x6. Second, by having the non-targetarrows turned in different directions or not. By having the non-target arrowsturned in different directions makes the task more difficult. This is consistentwith research that has shown that similarity between distractors (in this case thenon-upward pointing arrows) facilitates search (Duncan & Humphreys, 1989). It isgenerally more difficult to find the upward pointing arrow in the right matrix thanin the left matrix below. This means that the right-hand matrix requires longeraverage continuous glance times. 8 The S-IVIS task has actually been developed in connection with the European HASTE project.9 Cognitive psychologists Poulsen and Segalowitz have also explored motivation as a source forheightened schema activation. They conducted experiments to see how motivation affects attention(Poulsen & Segalowitz, 2000). By using a point system and auditory feedback these authors found thatmotivation manipulations selectively affected attention switching mechanisms, although basicresponse times were unaffected.

Right answer: ‘NO’Right answer: ‘YES’

INTRODUCTION

20

Figure 4: Examples of difficulty levels of S-IVIS tasks

Here, as with the simulator, the problems with choosing an experimentalsurrogate concern external validity, but the reliability of the distractor isconsidered more important in this experimental setting.

Using an ecological distractor: the glove box paper fall-out(experiment distractor II)As already mentioned there is an abundant amount of possible distractors that adriver in a modern car might be subjected to. Therefore, as a complement to the S-IVIS task, another distractor was proposed early in the planning of theexperiment. To distract the driver in a wholly different manner the sudden fall-outof paper from the glove box was used. This distraction event is very different fromthe S-IVIS distractor. In the way it was conceived, the glove box would suddenlyburst without the subject’s knowledge about it, thus come as a surprise. Thesubjects themselves are free to react however they want as opposed to the S-IVIStask where there are right and wrong answers.

That this distractor had weaknesses compared to the S-IVIS was concluded beforeexperimenting started. First, it is not possible to present it several times as adistractor, since the subject would soon learn that there is “not much to it”.Second, it does not require continuous attention. It is possible to chunk gaze timeon the distractor, or for that matter not look on it at all. It is, however, interestingto compare the results of this kind of distractor to the results of the S-IVISdistractor.

Using an experimental distraction criterionIn this study which only aims at finding out how effective two proposed distractioncountermeasures are, the distraction criterion could be rather conservative, that isnot allowing the subject to look away from the road much. This in order to providemore experimental data. It does not have to be realistic in the sense that thecriterion that is used in the experiment is a criterion that should be used in reallife. The criterion should not be too conservative either because then there is arisk that the countermeasures are activated too frequently, leading to frustrationof the subjects. Therefore, we choose a shorter time than 3.75 s that was used inthe Holmström and Johansson (2003) study, but a longer time than the averageglance times away from the road that were reported by the reviews of normalglance behavior cited on p. 10 (Green, 1999; Wierwille, 1993).

In the experimental setting, of the present study, there are predefined distractors(Experiment distractor I and II) which are located at the same position throughoutthe experiment. No other gaze directions than gaze on the road/main instrument

Easier Harder

INTRODUCTION

21

unit and gaze on the in-vehicle distractors are expected. It is therefore possible tomake the simplification to have only two zones; Zone 1, which corresponds to gazeon the road, and Zone 2, which corresponds to gaze inside the car.

In the terms of the distraction criterion model discussed on p. 14 the initial budgetis 3 s and the decrement constant 1. The increment constant will be set to infinity,which is a simplification, but no theoretical or other information was available toset this parameter more precisely.

Using steering reaction and glance duration times as dependentmeasuresThe countermeasures should be evaluated along the lines of how they redirect thegaze of the driver to the road, according to requirements 1 (direct gaze to the road)and 3 (fast) of the requirement list (p. 13). A dependent measure that could workas an evaluation tool in this respect is in-vehicle glance duration time. Bycomparing in-vehicle glance duration time when countermeasures are active to acontrol group it could be possible to evaluate the countermeasures in terms of howthey decrease in-vehicle glance duration time. The countermeasures are activatedaccording to the chosen distraction criterion of 3 seconds. This glance durationmeasure with a distraction criterion closely operationalizes the definition ofdistraction that we have and is made possible by the SmartEye eye-tracking tool.

To meet requirement 2 (enhance ability to cope with critical situation) of therequirement list, the subject should be exposed to a critical situation whiledistracted and warned. Since the simulator that is used in this study has beenvalidated for lane position (Harms, 1996; Törnros, 1998; Törnros, Harms, & Alm,1997), but not for collision avoidance maneuvers, a steering reaction time measureis proposed. The car is turned towards the ditch while the subject is distracted byan in-vehicle distractor (the S-IVIS). This forces the subject to steer back on theroad again with a swift maneuver to maintain lane position. See procedure of themethod chapter, p. 26, for details.

Using other elicitation techniquesThe dependent measures described so far may cover the three first items on therequirement list. To cover requirements 4 (unambiguous), 5 (not distract), and 8(accepted), a questionnaire was designed that the subjects completed after thetrial.

To acquire a possibility for a more qualitative analysis of reactions to the proposedcountermeasures, video recordings were also made during the trials.

METHOD

23

MethodThe experiment was carried out in the VTI simulator and a within groupcounterbalanced design was used with three conditions; CONTROL, FLASH, andBRAKE. As dependent measures a steering reaction and a glance duration timewere used. Other elicitation techniques for evaluation of the distractioncountermeasures included a questionnaire and DVD recordings.

Equipment

The VTI simulatorThe VTI driving simulator used in this study is made up of a visual system, amotion system and a sound system to simulate driving. The visual system ispowered by computer graphics and presents an image of the virtual road withthree projectors onto a 120 (width) x 30 (height) degrees screen. The motionsystem simulates acceleration motion with three degrees of freedom. Sideways +-3.5 m, pitch angle +15 degrees/-10 degrees, and roll angle +-24 degrees. Themotion system is also complemented with a vibration table which the cabin ismounted upon which primarily is used for simulating the vibrations caused by theroad-vehicle contact, but also for the simulation of small sudden movements. Thevibration table has four degrees of freedom +- 0.05 m vertically, 0.075 mlongitudinally, a pitch angle of +- 4 degrees and a roll angle of +-7 degrees. Forcesof the motion system on the cabin are downscaled to a half of the forces theysimulate. The sound system simulates road, wind and engine noise. The cabinitself was in this experiment a five-gear Volvo 850 that had been cut off justbehind the front seat with the front seats, dashboard, steering wheel, andeverything else in front of the B-column intact. The simulator logs data (e.g.steering wheel angle, speed, gear, etc.) at a maximum rate of 200 Hz.

The simulator can be used to simulate different kinds of driving environments likecity traffic, rural road traffic, etc. It is possible to simulate real roads in thesimulator; this has been done in validation studies of the VTI simulator (Harms,1996; Törnros, 1998; Törnros et al., 1997). These studies have primarily focused oncomparing the choice of speeds and lateral position for real driving and simulateddriving. As of yet, there have been no validation studies of more complex behaviortypes such as how people react in critical situations.

The S-IVIS task distractorFor the S-IVIS task a touch screen mounted just below the air intake panel wasused. In the modified version of the S-IVIS task used in this experiment, the textbuttons were translated into Swedish. The brightness was set to a low level partlybecause the subjects then had to “look harder” for the up-pointing arrows andpartly because it was easier for the experimenter to see what kind of picture thatcame on the screen via the sensitive monitoring camera that monitored how thesubjects pushed on the S-IVIS screen. For a thorough description of the S-IVIStask see p. 18.

METHOD

24

Figure 5: Experiment setup

The glove box distractorAnother distractor used in the experiment was the sudden outdepartment. The glove box door was locked by an electromagnet empty coke can wrapped up in a paper magazine. When thelectromagnet was turned off, the paper magazine worked likethe glove box door resulting in that the coke can together with down on the passenger seat beside the driver. This event was aeach subject and the subjects were not told about it before the tri

The Smart Eye system and IDPThe Smart Eye Pro Automotive 2.1 eye-tracking system wexperiment. Two cameras mounted on the dashboard and a cucomputer was part of the hardware. For this particular systefeatures had to be manually applied (by a lot of mouse clicking)subjects. These facial features were: left and right inner eyebrouter and inner eye-corner, left and right nostril, left and right and right ear and finally left and right eye center.

For this experiment seven snapshots were used to create the featracking: one facing straight ahead, two with the nose facing with the nose straight ahead but with gaze into each camera, subject looking at the S-IVIS screen just by slightly turning the hand another shot with the subject looking at the S-IVIS by bothand leaning forward, i.e. with the face closer to the screen. 14x7x2 (features x snapshots x cameras) possible features to ppositions for each subject. However, all features were not visibthereby mitigating the task for the experimenter who had to mmask”.

Together with the SmartEye system another software program,define two visual zones, a head-up zone (Zone 1) and a head-doThe border between these two zones was where the dashboard mwith an indent of Zone 1 in the dashboard where the speedomlooking at the speedometer activated Zone 1. Looking at t

SmartEye cameras

ba

a

s

om

teo

ll

h

S-IVIS screen

urst of the glovend inside was an

e current to the a spring to openthe magazine fell one-shot-only forl.

as used in thetomized personalm version, facialto pictures of thew, left and rightouth corner, left

ure mask for eye-ach camera, twone shot with theead to look down, turning the headThere were thusace in the propere on all pictures,ake the “feature

IDP, was used town zone (Zone 2).et the windshield,eter was so thate S-IVIS screen

METHOD

25

activated Zone 2. The borders of these two zones had to be defined individually foreach subject. In the figure below a subject is looking at the S-IVIS screen. Theupper window is the SmartEye window. The lower window is the IDP windowwith the two zones.

Figure 6: SmartEye screen with IDP

The Flash countermeasureTwelve blue diodes powered by a potentiometer were mounted at the dashboard sothat the blue light was reflected in the windshield approximately at middle of theroad position. The blink cycle was 0.3 s with 0.1 s lights on and 0.2 lights off. Alldiodes blinked synchronously. Total blink time was maximum 1.4 s and blinkingwas interrupted if Zone 1 was reactivated (i.e. if subjects looked up).

The Brake countermeasureThe brake countermeasure was a single pulse that was simulated by adeceleration in the visual system of 2.5 m/s2 during 2 s. The deceleration of themotion system was, however, smaller since the vibration table, the part of themotion system that can move at the speed which is required, had a limited pitchcapacity. The deceleration force exerted on the cabin by the vibration table was0.07 g.

METHOD

26

SubjectsThe subjects were recruited through the VTI subject pool which is comprised ofvolunteers that have seen ads in the local news-paper and at the local job center orheard via other people that VTI needs subjects. In this way 16 female and 14 malesubjects with an age from 18 to 52 years (mean 35,2, std 7,0) that did not wearglasses or make-up during the experiment were recruited. Glasses and make-upwere banned since the pilot study had indicated that eye-tracking was worse withglasses and make-up. Driving experience varied among the subjects, from thosewho had just received their license up to those who had held their license for morethan 30 years (mean 16,1 std 7,0). The subjects in the main study were numberedfrom 19 to 48 (lower subject numbers were used in the pilot study and as a buffer).

ProcedureThe subjects were given written instructions that informed them about that thestudy was about how drivers can manage other tasks while driving a car and thatdifferent situations put different demands on the driver. Certain situations likedriving in environments with a lot of other road users do not leave much room forother than the driving itself but at other times one should be able to carry outother things while driving, e.g. while driving in simpler environments such as on afreeway. Further, the subjects were instructed that only freeway driving wouldtake place in this experiment. On the other side of the written instructions page,instructions for the S-IVIS task were given (see appendix A).

After the written instructions had been read by the subjects, the same instructionswere repeated verbally by the experimenter making sure that the subjects hadunderstood the S-IVIS task. Then the simulator was shown for the subjects andfurther instructions on how to record a movie and taking some still pictures forthe personal profile that the eye-tracking system needed were given to thesubjects. The movie and the pictures were taken directly and the experimenterwho created the SmartEye personal profile (“feature mask”) could do so while thesubject practiced on the S-IVIS task (14 presentations) and practiced driving thesimulator for ten minutes. After the practice drive and when the profile and zoneshad been defined, the subjects were asked if they would like to have some coffee.This was offered since the pilots had shown that subjects easily got tired duringthe experiment. After these preparations the experiment began.

The road scenario was made as monotone as possible to make the driver willing toengage in the distraction task. The simulated road was a 110 km/h (~70 mph)freeway with no other traffic. One session lasted for an hour, 3600 seconds. Duringthis time 84 S-IVIS presentations in 12 blocks were made. Each presentationlasted for maximum five seconds. If the subject had not pressed a button by then itwas recorded as a “no answer” and the wrong answer beep sounded. The samething happened if the subject looked away from Zone 2 before pressing a button.During the entire trial the subjects were monitored with the SmartEye eye-tracking system.

In each block there were seven S-IVIS presentations of which one was extradifficult (6x6 matrix and arrows in all directions). If the subjects looked down formore than 2 seconds during such a difficult presentation the car wasautomatically turned, gradually during 1.5 s, 1.5 degrees in the direction of theright hand ditch. This forced the subject to turn the wheel in order to avoid ending

METHOD

27

up in the ditch. The yaw only took place if the subject looked down on the screen,or, to be exact, if Zone 2 was activated. If Zone 1 was activated, i.e. if the subjectlooked up again before 2 seconds, the yaw never set in. Moreover, if the subjectlooked up during the turn of the vehicle, the vehicle stopped turning, i.e. the yawwas interrupted. The time from yaw onset till the driver started to turn the wheelwith a sudden steering wheel movement was measured and called steeringreaction time. A sudden steering wheel movement was operationalized as steeringwheel angle change to the left of more than two degrees that had to be continuouswith an angle change rate of more than 0.020 degrees for each time frame (0.005s). This criterion was set by analyzing the data from the pilot experiments.

All the S-IVIS presentation glance times from the start of the S-IVIS presentationto that the subject looked up, in other words, from Zone 2 activation to Zone 1reactivation were measured. The glance time at difficult presentations was calledglance duration time at difficult S-IVIS presentations.

During the experiment all of the subjects were subjected to three conditions –FLASH, BRAKE and CONTROL – each of which lasted for twenty minutes. Theorder of the conditions were counterbalanced between the 30 subjects and sincethere are 6 possible orders each condition order was replicated 5 times. Thedistraction criterion in the experiment was three seconds which means that thesubjects were categorized as distracted if Zone 2 had been activated for more thanthree seconds. This means that during the twenty minutes that the FLASHcondition lasted the diodes in the windshield started to flash after that Zone 2 hadbeen activated for more than 3 seconds. When the subjects drove in the BRAKEcondition the brake pulse was activated after 3 seconds of Zone 2 activation. In theCONTROL condition nothing happened when the driver met the distractioncriterion.

Below is a schematic diagram of the hypothesized chain of events in theexperimental setup at difficult S-IVIS presentations and how the differentreaction times are related to these events and to each other. A third concept hasalso been added to the diagram and that is “visual processing time”, the time ittakes after looking up before course heading is adjusted.

Figure 7: Hypothetical relation between variables

Subj. starts steering

Predefined:

Visual processing time

Yaw Count.meas. Subj. looks up

2 s 1 s

Glance duration time

Steering reaction time

Subj. looks down

METHOD

28

At the end of the trial, at 3505 s, the glove box burst and glance behavior for thatdistraction was registered. After the trial, the subjects filled out a questionnaire,filled out the pay check form (each subject received SEK 800 ~ 100$ or 90€) andwere debriefed.

During the entire trial the subject was monitored visually via the SmartEyecameras, a subject monitoring camera, and a camera showing the S-IVIS screenand how the subject pressed the S-IVIS screen buttons. The simulated road scenethat the subject saw was also presented to the experimenters on three separateLCD computer screens. The experiments could also hear via a microphone whatwas going on in the simulator cabin and could also talk to the subject via intercom.The subject monitoring camera picture, the S-IVIS camera picture, and the centerroad scene picture were recorded on a three field DVD.

Pilot experimentBefore the actual experiment took place a pilot experiment was conducted to testthe procedure and the equipment and to see if the S-IVIS distraction task wasdistracting enough. As a result of the pilot experiment the feedback sound wasadded to the S-IVIS task and the yaw angle lowered from 2 to 1.5 degrees and therule that when the subject looks up the yaw is interrupted. The criterion of asudden steering wheel was also set by data from the pilot experiments. It was alsonoted that, by the reactions of the pilot experiment subjects, that thecountermeasures could have effect.

RESULTS

29

ResultsThe distraction criterion was met 123 times during the CONTROL condition, 162times during the FLASH condition, and 134 times during the BRAKE conditionfor all subjects during the entire experiment. Corresponding means are hence(n=30), 4.1, 5.4, and 4.5. At difficult S-IVIS presentations the criterion was met 41,44, and 37 times respectively (means: 1.37, 1.47, and 1.23). Five subjects had morethan half (229) of the total distraction events (419). The five subjects who hadleast distraction events only shared 1 distraction event. Only seventeen subjectswere distracted in every condition. During the whole experiment drive that lasted3600 s the mean eye-tracking fall-out for all subjects was 88.1 s (std=95.2) or2.45%, ranging from 0.09% to 10.2% for the individual subjects (see appendix C-1).The first two subjects’ results for the beginning of the trial (28 first S-IVISpresentations) were disregarded in all of the calculations of the data since the eye-tracking equipment was extra unstable at that time.

Glance duration time after conditionThe mean glance duration time at difficult S-IVIS presentations, on occasionswhen the distraction criterion was met (i.e. gaze down more than 3 seconds), wasfor the CONTROL condition 3.87 s (std=0.56), for the FLASH condition 3.86 s(std=0.44), and for the BRAKE condition 3.76 s (std=0.52). The mean glanceduration times from look down to look up for all S-IVIS presentations on occasionswhen the distraction criterion was met was for the CONTROL condition 3.65 s(std=0.59), for the FLASH condition 3.68 s (0.62) and for the BRAKE condition3.59 s (std=0.52). The high number of missing data, with only seventeen subjectsbeing distracted in every condition, and the large variances compared to the meandifferences between the conditions, prohibits and makes it unnecessary to carryout inferential statistical tests.

Figure 8: Glance duration results

��������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������

Glance duration time at all S-IVIS

0

0,5

1

1,5

2

2,5

3

3,5

4

4,5

CONTROL FLASH BRAKE

Condition

Gla

nce

dura

tion

(sec

onds

)���������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������������������

Glance duration time at difficult S-IVIS

0

0,5

1

1,5

2

2,5

3

3,5

4

4,5

CONTROL FLASH BRAKE

Condition

Gla

nce

dura

tion

(sec

onds

)

RESULTS

30

Steering reaction timeOf the 122 distraction events at difficult S-IVIS presentations, the steeringreaction criterion was met 103 times; 36 times in the CONTROL condition, 36times in the FLASH condition, and 31 times in the BRAKE condition. The meansteering reaction time on these occasions was measured from yaw onset to that thesubject started a sudden steering movement. This reaction time was 1.05 s(std=0.38) for the CONTROL condition. For the FLASH condition it was 1.10 s(std=0.51), and for the BRAKE condition it was 1.08 s (0.42). Here, as well asbefore, the high number of missing data and the large variances compared to themean differences rule out an inferential statistical test.

Figure 9: Steering reaction results

Average visual processing time (steering reaction time – (glance duration time – 2s)) is thus for the CONTROL condition –0.82 s, for the FLASH condition –0.76 s,and for the BRAKE condition –0.68 s.

An example of steering data and glance data for an individual distraction event ispresented in appendix C-3.

Glance duration times overallThe mean glance duration time of all 2,413 glance times on the S-IVIS for allsubjects (n=30) was calculated to 2.11 s and the median to 1.97 s. Below is thedistribution of glance durations shown.

����������������������������������������������������������������������������������������������������������������

����������������������������������������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������

Steering reaction time at difficult S-IVIS

0

0,5

1

1,5

CONTROL FLASH BRAKE

Condition

Stee

ring

reac

tion

time

(sec

onds

)

RESULTS

31

Figure 10: Glance duration result histogram

18 % of the total number of 2,413 glances were longer than 3.00 seconds. Lessthan 5 % of the total number of glances on the S-IVIS were longer than 3.82 s (95th

percentile). The 97.5th percentile was 4.30 s and the 99th percentile was 4.84 s. Theaverage glance times for the difficult S-IVIS was 2.61 s. The means for every S-IVIS presentation is presented in appendix D-2. The standard deviations for theglance durations for each of the 84 presentations were calculated and the mean ofthese standard deviations was found to be 0.84 s. The mean of the standarddeviations for the difficult S-IVIS was 1.03 s.

Average glance times after gender and ageThe average of the average glance times for all females (n=16) was 1.76 s, thecorresponding figure for males (n=14) was 2.56 s, the measured gender differencewas thus 0.80 s or 45.3% of the lower value. Homogeneity of variance was testedand it was found that there was no significant difference in variance between thetwo groups F(13,15)=1.81, ns. A two-tailed t-test, assuming homogeneity ofvariance, revealed a significant difference between the means t(28)=4.91, p<0.01.The confidence interval around this observed difference is CI99=0.80±0.45, i.e. with99% confidence we can say that the difference between females’ and males’ glancetimes lies between 0.35 and 1.25 s.

The subjects were grouped into three age groups: 18-30, 31-40, and 41+. Theaverage of the average glance times for subjects between 18-30 (n=6) was 2.06 s,the average for the 31-40 year-olds (n=18) was 2.22 s, and for the 41+-group (n=6)the average was 1.94 s. An ANOVA test showed that there were no significantdifferences between these three groups F(2, 27) = 0.53, ns.

S-IVIS answersTo see how well subjects answered the S-IVIS task, the 84 S-IVIS presentationanswers were summarized. The average number of correct answers for all subjects

020406080

100120140

0 1 2 3 4 5 6

Glance duration (seconds)

Num

ber o

f gla

nces

0%

20%

40%

60%

80%

100%

Cum

ulat

ive

fract

ion

RESULTS

32

was 60.8 (std=10.3), the average number of wrong answers 6.4 (std=4.0) and theaverage number of “no answer” 16.8 (std=10.7). The mean for females/males was58.3/63.7 (correct answer), 5.8/7.1 (wrong answer), and 19.9/13.1 (no answer). Astatistical test was carried out to see if there was a significant difference betweenthe sexes in number of correct answers. Homogeneity of variance was tested and itwas found that there was no significant difference in variance between the twogroups F(13,15)=1.99, ns. A one-tailed t-test, alpha = 0.01, assuming homogeneityof variance, testing the null-hypothesis that the females did not have less correctanswers showed that there was no significant difference between the meanst(28)=1.45, ns.

Glove box distractor glance behaviorA general characteristic for glove box distractor glance behavior was that glancetime off the road was chunked, i.e. the subjects looked back and forth between theroad and the passenger seat. The in-vehicle glances that were not separated bymore than 10 seconds of on the road glances were categorized as being a part ofthe reaction to the glove box distractor. Mean in-car single glance length was 0.62s (std=0.37) for those glances. Mean number of glances was 2.27 (std=1.39). For allsubjects’ glove box distractor glance behavior see appendix D.

QuestionnaireThe subjects responded to the questions in the questionnaire by checking one ofseven boxes. The leftmost box was labeled “not at all” and the rightmost box“very”. In the data analysis, the “not at all”-category corresponds to 1 and “very” to7.

The results of some selected questions about what the subject thought of thecountermeasures were:

Item Question CM Median Mean Std

12d

What did you think of the flashingblue light?

Draws attention upon itself FLASH 6 5.0 1.9

12e Draws attention upon the road FLASH 4 4.1 1.8

10d

What did you think of the strongbrake pulse?

Draws attention upon itself BRAKE 5 4.6 1.5

10e Draws attention upon the road BRAKE 5 4.8 1.5

13 Do you think that the flashing bluelight would be a good part of an anti-distraction system in cars?

FLASH 4 4.4 1.8

11 Do you think that the strong brakepulse would be a good part of ananti-distraction system in cars?

BRAKE 5 4.5 1.6

Table 2: Questionnaire results

RESULTS

33

To see if there was a difference between how subjects answered question 12d and12e a Mann-Whitney U test for large samples (Hinkle, Wiersma, & Jurs, 1998)was carried out which showed that there was no significant difference between thesubjects’ answers U=310, z=0.21, ns.

One question asked about how much money the subjects would be willing to spendon an anti-distraction system. The answers varied from the equivalent of 0€ to2,250€, the median was 100€. The medians, means and standard deviations of allthe questionnaire items are presented in appendix B.

Video recordingsThe video recordings were used to analyze the effect of the kinesthetic brake pulsefor every time (134 times) it was activated. The mean glance duration time forevery category was also calculated.

Category Observed behavior Number ofevents

Mean glance duration time

1 No reaction to brake pulse 9 4.55

2 The subject pushes S-IVIS button,then the brake pulse comes butdifficult to say if it is the brake pulsethat makes the subject look up

32 3.49

3 The subject looks up briefly becauseof the brake pulse, but down again totry to solve S-IVIS task

25 3.95

4 The subject seems alerted by brakepulse

3 3.92

5 The brake pulse has clear effect 7 3.85

NO The brake pulse is not visible onvideo

11 3.45

LATE The subject has looked up before thebrake pulse comes

47 3.24

Table 3: Video results

Category 5 “The brake pulse has clear effect”, means that the subject feels thebrake pulse and as a reaction looks up on the road. Category 3 is similar to 5, butwith the difference that the subject does not keep gaze on the road as a reaction tothe brake pulse, but looks down again.

DISCUSSION

35

DiscussionThe aim of this study was to evaluate two proposed countermeasures and exploremethodological techniques for doing so. A requirement list was made (p. 13) thatwas to guide the evaluation of the countermeasures. We said that the flash andbrake countermeasures a priori fulfilled requirements 6 (not occlude), 7 (notauditive), and 10 (realizable) of this list. We then said that it requires empiricalinvestigation to test the other requirements. An experiment was thus designed tosee whether the countermeasures directed gaze to the road (1), enhanced ability tocope with a critical situation (2), whether they were fast (3), unambiguous (4),undistracting (5), and accepted (8). That the countermeasures were reliable (9)after long periods of usage and despite large variations in the ambientenvironment was not made the objective of the study.

A glance duration measure was used to address the first and third issues. Asteering reaction measure was used to address the second issue. A combination ofquestionnaire questions and the quantitative measures was used to acquireknowledge about the countermeasures in respect to requirements 4, 5, and 8.

The results show that neither of the proposed countermeasures contributed tolower the glance duration away from the road or lower the steering reaction time.According to the experiment, requirements 1, 2, and 3 were therefore not fulfilled.There was a very slight tendency that the BRAKE condition produced lower timesthan the FLASH and the CONTROL group in both quantitative measures, butthis tendency is statistically insignificant. The subjects’ answers to the twoquestions about how much the flash drew the attention upon itself and how muchattention it drew upon the road was 6/7 and 4/7 respectively. This difference wasnot significant, but the tendency might suggest that the flash countermeasure didnot fulfill requirements 4 (unambiguous) and 5 (undistracting). For the brakecountermeasure there was no such difference (5/7 for both questions) which meansthat the questionnaire could not show that the brake countermeasure wasambiguous or distracting. The subjects had a slight tendency to think better of thebrake pulse than of the flash also when they were asked whether thecountermeasure would be a good part of an anti-distraction system in cars(Flash=4/7 and Brake=5/7). This suggests that compared with the flash, the brakepulse was a slightly better alternative. However, the difference is very slight andfar from significant. The subjects ratings in this issue were not convincing whichmeans that the question whether requirement 8 (accepted) has been met remainsopen.

The groups that had a significant difference in glance times were not the groups ofthe different conditions. The only difference that was found was the differencebetween females and males in overall off-road glance time. This is interestingbecause it suggest that one of the sources for higher male crash involvement, asdescribed elsewhere (Al-Balbissi, 2003), may have to do with attention behavior.

That no significant differences for steering reaction time and glance duration timewere found between the treatment groups and the control group may eitherdepend on an actual lack of efficiency of the countermeasures or on methodologicalshortcomings. Both these alternatives will now be discussed.

DISCUSSION

36

Inefficient countermeasuresThe countermeasures that were chosen were chosen because previous researchhad indicated that there is a possibility that they would be effective. A brake pulsehad been used by Shutko (1999) and Dingus (1996). Their experiments were,however, on-the-road studies and so they could use a strong brake pulse (0.3 g and0.2 g respectively). The motion base of the simulator that was used in this studyonly provided a limited brake force that was much lower (0.07 g) than the oneused in previous research. Although the pilot experiment had indicated that thebrake pulse might have effect, it was maybe too weak for showing an effect in thedependent measures used in the study. If intensified, it might perhaps show moreeffect. The video analysis category “NO” (11 occasions) also suggests that thebrake pulse was too soft. In the video analysis, on occasions when the brake pulsewas categorized as having clear effect, the average glance time was not muchlower (3.85 s) than the CONTROL group average glance time (3.87 s). This mayalso indicate that the glance duration measure used in this study is notappropriate for evaluating brake pulses. We will soon discuss the glance durationmeasure more below.

The previous research that was an argument for using a flash as acountermeasure was the results of experiments that had shown that suddenvisual onsets capture attention (Remington et al., 1992; Yantis & Hillstrom, 1994).These results were, however, based on laboratory experiments and only slightdifferences in response times due to interference of stimuli in non-target positionswas the basis for the authors to claim that sudden visual onsets capture attention.The character of the conditions in the present study was quite different, andtransfer of the Remington et al. effect to the conditions of the driver distractioncontext was perhaps only limited. That there was a tendency that the subjectsthought the flash drew more attention to itself than to the road also suggests thatthe flash as it was designed in this experiment is not optimal as acountermeasure.

The discussion whether or not to allow auditory signals should also come up again,since any effect of the brake pulse and the flash could not be shown. Since manyother studies have fruitfully used auditory signals as warning signals, and a largepart of drivers do have hearing intact, the seventh requirement, not auditive, ofthe countermeasure requirement list should perhaps be excluded.

Methodological shortcomingsThe most important elicitation techniques that were used in the experiment wereeye-tracking and simulator logging, which rendered the two measures of glanceduration and steering reaction. Neither of these two measures has beensatisfyingly validated before the experiment. We will now discuss, in light of theresults of the experiment, what the flaws of these measures were and what isneeded to develop them. The derivative of glance duration time and steeringreaction time – visual processing time – will also be discussed. Then five otherpossible error sources will be discussed.

Glance duration timeMany studies have studied driver glance behavior (Green, 1999; Harbluk et al.,2000; Land & Lee, 1994; Wierwille, 1993). The automatic and unobtrusive kind of

DISCUSSION

37

eye-tracking that was used in this study has, however, not been so widely usedyet. This technology has, nevertheless, a good potential for becoming a goodresearch tool and driver attention support system component in future cars.Enhancements must, however, be made to and thorough validation be carried outof the SmartEye system before it is further used as a research tool, to ensure exactmeasurement.

The average eye-tracking fallout rate of 2.45% is far too high, especially when oneconsiders the fact that it is at the most crucial moments, when the driver changesgaze and head direction, that fallout is as biggest. It is suspected that theSmartEye system has a lag, i.e. that it does not follow gaze as closely as would bedesired. The automaticity of the system must also be increased to increasereliability. First, the feature mask must be automatically generated. Second, thezones must be the same for all drivers, i.e. when two different drivers look at thesame point in the driving environment the resulting measurement output shouldbe the same. This second point has to do with that the coordinate system shouldnot be person based, but vehicle based. To increase precision of the glanceduration measure, the fixations should be possible to measure so that in-vehicleglance duration time is measured from the last fixation on the road before lookdown to the first fixation on the road after look up. In this study the time startedand stopped somewhere in between fixations on the road and on the distractors.

A more rigorous evaluation of the SmartEye system than that carried out byHolmström & Johansson (2003) is needed. Comparisons of the SmartEye systemmay be made to a competing system on the market produced by SeeingMachines.This system has undergone a partial validation (Victor, Blomberg, & Zelinsky,2001) of the kind that is urged for with the SmartEye system. To validate theexact measurement of the SmartEye system would need a set of experiments andpossible enhancement iterations of its own.

We mentioned above that the glance duration measure might not at all be ideal toevaluate distraction countermeasures. When the video analysis showed that thebreak pulse had clear effect the average glance duration was not much smallerthan for the CONTROL group. This may be due to the aforementioned lack ofprecision in the eye-tracking equipment, but could also be due to something else.

Steering reaction timeThe steering reaction criterion was met on 84% (103/122 occasions) of the difficultS-IVIS presentations when the subjects reached the distraction criterion. Here,fallout depends largely on that the subjects made steering wheel movementsbefore the turn set in or had a high steering wheel angle variability, making theyaw seem small. It could even be so that the artificial yaw resulted in that the carwas turned into lane, in cases when the car was heading to the left just before theyaw. The definition of a sudden steering wheel movement may also be revised.Fallout was largely due to the character of the yaw, but it may also be the casethat the criterion resulted in that too many steering reactions were reported. Inone calculation the criterion was changed so that only steering wheel movementsthat were more than 3 degrees were recorded10. This did, however, not lead tofewer false alarms but rather to more misses. In the steering reaction criterion 10 It still had to be continuous with an angle change rate of more than 0.020 degrees for each timeframe (0.005 s).

DISCUSSION

38

that was chosen for this experiment it was the continuity of the movement and thetotal size of the movement which were most emphasized. The swiftness of motionwas less weighted since a steering wheel speed of 4 degrees/second sufficed tofulfill the requirement.

The validity of the steering reaction measure may also be questioned since therequired turning of the car towards the ditch when the driver looks inside the caris a rather artificial intervention. However, it may have revealed interestingaspects of vision of distracted drivers that have to do with peripheral vision. Thisis what will be discussed next.

Visual processing timeIt was anticipated that the subjects would look up before they started their suddensteering wheel movement away from the ditch. The result was, however, that mostsubjects started to steer away from the ditch before they looked up. The followingfigure is a modification of Figure 7 following these results.

Figure 11: Result relation between variables

There could, at least, be three reasons to why the visual processing time wasnegative in the experiment. It could be due to lag of the eye-tracking system, itcould be due to that the steering reaction measure criterion was too sensitive, or,more interestingly, it could be explained by that steering is sufficiently subservedby peripheral vision. The two first sources of this deviation from the hypothesishave already been discussed. Let us see what the third means.

Is it possible that the participants in the study are able to react to the artificialyaw while looking at the in-vehicle distractor? There are indeed studies thatsuggest that detection and spatial orientation may only require peripheral and notfoveal vision to work. A study by Lamble, Laakso, & Summala (1999) “showedmodest losses in drivers’ detection abilities, in terms of TTC, in closing headwaysituations when the driver’s visual attention was focused away from the roadscene ahead.” (p. 814). Their subjects could react to a decelerating car ahead bybraking while focusing gaze on in-vehicle tasks inside the car.

If steering is regarded as a spatial orientation task the following passage fromLeibowitz (1986) is interesting: “because spatial orientation can be adequately

Subj. starts steering

Predefined:

-Visual processing time

Yaw Count.meas. Subj. looks up

2 s 1 s

Glance duration time

Steering reaction time

Subj. looks down

DISCUSSION

39

subserved by low spatial frequencies, visual guidance is quite adequate in theperipheral fields (Leibowitz, Post, Brandt, & Dichigans, 1982). This can beillustrated by our ability to walk or drive a vehicle when vision is blurred or with amask covering the central field.” (p. 605). This could mean that we may look at anin-vehicle location and still keep lane and may therefore explain how a negative“visual processing time” is possible. The deception of artificially yawing the carmay also still work since “spatial orientation is carried out reflexively withminimal awareness” (ibid. p. 605), which could explain why the subjects are notaware of what made them turn the wheel.

Missing dataThere were altogether 2520 (84x30) S-IVIS presentations, but only 2413 (96%)glances into zone 2 during these presentations. 56 of the missing glances areexplained by subjects 19&20:s fallout in the control condition. The remaining 51non-look down cases, which are probably mostly due to eye-tracking problems, aredistributed among 11 subjects. Of the 360 (12x30) possible glance duration datapoints at difficult S-IVIS are only 122 (34%) over 3 s. Mean glance time was 2.61 s.This is probably mostly due to that the distraction criterion was high and that thedifference in difficulty between difficult and non-difficult S-IVIS presentationswas not large enough. Of the 122 possible steering reaction data points, when thesubjects were categorized as distracted at difficult S-IVIS, only 103 (84%) wererecorded. This is probably mostly due to that the steering reaction criterion had itsflaws (see above) and that the yaw was not large enough.

All the missing data described above, together with the small glance time meandifferences between the conditions, resulted in that an ANOVA test was notcarried out. There were only seventeen subjects who were distracted in allconditions and in order to be able to use a dependent samples ANOVA, data fromall subjects must be available.

Nuisance warningsThe effect of the countermeasures could have been lessened by the many falsealarms. There were in total 296 activations of the countermeasures. 81 of theseoccurred when the vehicle was turned towards the ditch, i.e. when the driver wassubjected to an actual close call. On the other 215 occasions, the countermeasureswere activated without anything dangerous or close callish occurring in theenvironment. Since nothing dangerous occurred in the environment at those 215occasions, the countermeasure activations could be interpreted as false alarms ornuisance warnings. Repeated nuisance warnings may lead to that warningseventually will be ignored. This could have been the case in our experiment.

Insufficient distractorsIf the two glance duration histograms that have been presented previously (on pp.9 and 30), are compared it is clear that the S-IVIS results in longer glances off theroad than was the case in the Hada study (1994). The median glance time off theroad for all S-IVIS of 1.97 s is, however, comparable to the study by Kimura et al.(1990) that had medians up to 2.0 s. The average glance time for the difficult S-IVIS of 2.61 s is considerably higher than what is reported by any of the glancetime sources that were presented in the introduction. The S-IVIS distractor waschosen because it was said that it might have the ability to continually distract

DISCUSSION

40

most subjects for a relatively long time. The relatively high glance duration meansand low standard deviations suggest that the S-IVIS is a fairly good distractor inthat respect. However, it was not enough in combination with other experimentalflaws to distract the subjects reliably.

If the simplification that gaze is equal to attention does not hold, then the S-IVISmight perhaps not be a distractor at all. This reasoning is supported by thenegative visual processing time discussed above. If the drivers were not reallydistracted in the sense that they did not take their mind off the road, couldexplain why there was a lack of significant results in the steering reactionmeasure.

The results of the glove box distractor were much different from the S-IVISresults. They are more similar to glance behavior that is described in Wierwille(1993). The variance between subjects, how many glances and how long theseglances are makes it clear that the glove box distractor is not useful whenextended continuous glance times are wanted.

Non-precise hypothesesIt is not only the countermeasures per se or the used methodology that have flaws.We may conclude from the discussion up to now that the hypothesis formulationwas flawed. If the hypothesis formulation had been more precise (for instancedetailing the force of the brake pulse), perhaps some of the problems of the studywould not have occurred. However, at this stage of the research process, thingshave to be discovered and hopefully the hypotheses of future studies on thissubject will be more acute.

Lack of separation between the effect of the countermeasures andthe psychological force to look on the roadThe psychological force to look on the road was described above (p. 9) and it wasconcluded that it is very rare that subjects look away from the road for longerintervals of time. What this study wanted to investigate was, however, behavior ofsubjects who do look away from the road for longer intervals of time and to seewhat could be done to redirect their gaze to the road again. It was thereforenecessary to expose the subjects to a task, which resulted in extreme glance timesoff the road. Then the next step was to measure the effect of a countermeasurethat redirects the gaze to the road again. However, when gaze has been off theroad for such a long time it could be the psychological force to look on the road andnot the countermeasure that makes the subject look back. To separate the effect ofthe countermeasures from the psychological force to look on the road is thereforedifficult. This may be the most important explanation to why there were nodifferences between the treatment groups and control group in this experiment.

With more reliable eye-tracking, a lower difficulty level of the non-difficult S-IVIStasks, which would lead to less false alarms, and a lower distraction criterion itwould perhaps be possible to see an effect of the brake pulse countermeasure. Butthat result would still be contaminated by the psychological force to look on theroad.

DISCUSSION

41

A comprehensive attention support systemWe have now discussed the possible inefficiency of the countermeasures and thepossible shortcomings of the methodology of this study. It remains, however, todiscuss the distraction countermeasure idea. To solve driver distraction problemswith a distraction countermeasure system, and a driver distraction criterion, asthey have been defined in this thesis (p. 11 and 14), is perhaps not a viablesolution. To have a driver distraction criterion expressed in seconds without usingany other information about the driver-vehicle-road system is questionable. Fartoo many false alarms or a uselessly liberal distraction criterion would be theresult of such a system. It would perhaps be better if sensor information about thedriver’s gaze is used as a complement to other sensor information about thedriver-vehicle-road system. Examples of which parameters other sensors couldgather information about is: drowsiness, head-way, time to collision, lane position,speed, road type, etc.

A comprehensive attention support system could perhaps be made up of thefollowing parts: a) sensor-technology; b) criteria for what sufficient attention iswhich must be adaptive to vehicle capability, driver capability, drivingenvironment, and effector capability; and c) effector-technology.

Figure 12: Components of an attention support system

Research questions that are made topical include: Which information about thedriver-vehicle-road system do we need to algorithmically assess whether thesystem is in a state of hazardous distraction?; How can driver status, environmentstatus, and vehicle status be measured?; Which measure/s are mostreliable/feasible?; What should be the distraction criterion? (with this question thesignal detection problem is brought to the fore, i.e. the trade off between falsealarms, and misses, of a supposed attention criterion); and finally, Whichrealerting effectors are effective?

All of these questions must be seriously addressed before an attention supportsystem is properly realized. In this master thesis the focus has been on the latterof the questions, i.e. the evaluation of possible effector-technologies, although fullydeveloped sensing technologies and algorithms were not available in the project. Itwould perhaps have been better if the fact that step a and b were not fulfilled hadbeen recognized at an earlier stage in this project. As it were now, unfinished,non-validated technology and algorithms, were used for steps a and b. The resultswould perhaps had been better if the effector technology research of this thesiseither had had proper sensor technology and criterias, or had been completelydecoupled from the driving situation.

SensorsVehicle statusDriver statusEnvironment

AlgorithmCriterion

EffectorsCountermeasure

CONCLUSION

42

ConclusionAlthough no effects of the proposed driving distraction countermeasures could beshown, has this thesis:

• Integrated the driving theory of Gibson & Crooks into a driver distractiondefinition.

• Presented a requirement list for realerting stimuli (that also could apply to acomprehensive attention support system).

• Collected data in a simulator environment of responses to and glance behaviorconnected to a surrogate in-vehicle information system.

• Showed that there is a difference in in-vehicle glance time between femalesand males in the experiment set-up used, which is consistent with the greatercrash involvement of males.

• Suggested that people do not always first look on the road before they initiatesteering maneuvers.

In the future it is recommended that research should focus on comprehensiveattention support systems, integrating information from different sensors and notonly rely on eye-tracking information. Before eye-tracking equipment is used itshould be thoroughly validated and have a very low fall-out rate. If kinestheticbrake pulses are to be used as realerting stimuli, the force that they exert shouldnot be too weak.

This is what can be concluded with this thesis, and hopefully it has contributedwith knowledge as to mitigate the effects of driver distraction so that the numberof accidents on the world’s roads can be reduced.

REFERENCES

43

References

Al-Balbissi, A. H. (2003). Role of gender in road accidents. Traffic injuryprevention, 4, 64-73.

Dember, W. N. (1960). Psychology of Perception. New York: Holt, Rinehart andWinston.

Dingus, T. A. (1996). Performance guidelines for front-rear-end crash avoidancesystems based on phase II experimentation and recentresearch.Unpublished manuscript.

Duncan, J., & Humphreys, G. (1989). Visual search and stimulus similarity.Psychological Review, 96(3), 433-458.

Gibson, J. J., & Crooks, L. E. (1938). A theoretical field-analysis of automobiledriving. American Journal of Psychology, LI(3), 453-471.

GM. (2003). GM Ability -Safety - Milestones in driver distraction. Retrieved July15, 2003, fromhttp://www.gm.com/company/gmability/safety/senseable/milestones/milestones.html

Graham, C. H. (1965). Discriminations that depend on wavelength. In N. R.Bartlett, J. L. Brown, Y. Hsia, C. G. Mueller & L. A. Riggs (Eds.), Visionand visual perception (pp. 350-369). New York: John Wiley & Sons, Inc.

Green, P. (1999). Visual and task demands of driver information systems(Technical Report No. UMTRI-98-16). Ann Arbor, MI: The University ofMichigan, Transportation Research Institute.

Hada, H. (1994). Drivers' visual attention to in-vehicle displays: effects of displaylocation and road types (Technical Report No. UMTRI-94-9). Ann Arbor,MI: The University of Michigan, Transportation Research Institute.

Harbluk, J. L., Noy, I. T., & Eizenman, M. (2000). The impact of internaldistraction on driver visual behaviour. Retrieved June 15, 2003, fromhttp://www-nrd.nhtsa.dot.gov/departments/nrd-13/driver-distraction/Papers.htm

Harms, L. (1996). Driving performance on a real road and in a driving simulator:results of a validation study. In A. G. Gale & al. (Eds.), Vision in vehicles(pp. 19-26). Amsterdam: Elsevier.

Hinkle, D. E., Wiersma, W., & Jurs, S. G. (1998). Applied statistics for thebehavioral sciences. Boston: Houghton Mifflin Company.

Holmström, J., & Johansson, J. (2003). System för föraruppmärksamhet,Systemutvärdering. Unpublished Bachelor thesis, Högskolan TrollhättanUddevalla, Trollhättan, Sweden

James, W. (1890/1981). The principles of psychology. Cambridge, MA: HarvardUniversity Press.

Kimura, K., Osumi, Y., & Nagai, Y. (1990). CRT display visibility in automobiles.Ergonomics, 33(6), 707-718.

Lamble, D., Laakso, M., & Summala, H. (1999). Detection thresholds in carfollowing situations and peripheral vision: implications for positioning ofvisually demanding in-car displays. Ergonomics, 42(6), 807-815.

REFERENCES

44

Land, M. F., & Lee, D. N. (1994). Where we look when we steer. Nature, 369, 742-744.

Landau, F. H. (1995). Human factors design of collision warning systems. Paperpresented at the Second world congress on intelligent transportationsystems, Tokyo, Japan.

Lee, J. D., McGehee, D. V., Brown, T. L., & Reyes, M. L. (2002). Collision warningtiming, driver distraction, and driver response to imminent rear-endcollisions in a high-fidelity driving simulator. Human Factors, 44(2), 314-334.

Leibowitz, H. W. (1986). Recent advances in our understanding of peripheralvision and some implications. Paper presented at the 30th annual meetingof the Human Factors Society.

Leibowitz, H. W., Post, R. B., Brandt, T., & Dichigans. (1982). Implications ofrecent developments in dynamic spatial orientation and visual resolutionfor vehicle guidance. In A. W. Wertheim, W. A. Wagenaar & H. W.Leibowitz (Eds.), Tutorials in motion perception: Plenum.

Mackie, J. L. (1965). Causes and Conditionals. American Philosophical Quarterly,2, 245-265.

McKnight, A. J., & Adams, B. B. (1970). Driver education task analysis: Taskdescriptions, 1. Alexandria, VA: Human Resource Organization.

Merat, N. (2003a). Loading Drivers to their Limit: The Effect of IncreasingSecondary Task on Driving.Unpublished manuscript.

Merat, N. (2003b). Personal communication.Michon, J. A. (1985). A critical view of driver behavior models. What do we know,

what should we do? In L. Evans & R. Schwing (Eds.), Human behaviorand traffic safety (pp. 485-525). New York: Plenum Press.

Norman, D. A., & Shallice, T. (1986). Attention to action: willed and automaticcontrol of behaviour. In R. J. Davidson, G. E. Schwartz & D. Shapiro(Eds.), Consciousness and self-regulation: Advances in research and theory(Vol. 4, pp. 1-18). New York: Plenum.

Norton, R., Hyder, A., & Peden, M. (2001). The epidemiology of road trafficinjuries: what exists, gaps and the role of public health. In M. M. Peden(Ed.), Proceedings of WHO meeting to develop a 5-year strategy for roadtraffic injury prevention (pp. 14-23). Geneva: WHO.

NTSB. (2002). NTSB Highway accident brief, accident no. HWY-98-FH-029.Retrieved July 17, 2003, fromhttp://www.ntsb.gov/publictn/2002/HAB0216.htm

Onken, R. (1994). DAISY, an adaptive, knowledge-based monitoring and warningsystem. In Proceedings of the 1994 Vehicle Navigation & InformationSystems Conference (pp. 3-10). Yokohama, Japan: Institute of Electricaland Electronics Engineers.

Peden, M. M. (Ed.). (2001). Proceedings of WHO meeting to develop a 5-yearstrategy for road traffic injury prevention. Geneva: WHO.

Poulsen, C., & Segalowitz, N. (2000). Selective effects of prior motivationalexperience on current on-line control of attention. Brain & Cognition, 43(1-3), 365-370.

Remington, R. W., Johnston, J. C., & Yantis, S. (1992). Involuntary attentionalcapture by abrupt onsets. Perception & Psychophysics, 51(3), 279-290.

Rensink, R. A. (2000). When good observers go bad: change blindness,inattentional blindness, and visual experience. Psyche, 6(9).

REFERENCES

45

Roberts, I., Mohan, D., & Abbasi, K. (2002). War on the roads: the public healthcommunity must intervene. British Medical Journal, 324(7346), 1107-1108.

Senders, J. W., Kristofferson, A. B., Levison, W. H., Dietrich, C. W., & Ward, J. L.(1967). The attentional demand of automobile driving. Highway researchrecord, 195, 15-33.

Shutko, J. (1999). An investigation of collision avoidance warnings on brakeresponse times of commercial motor vehicles [Electronic version].Unpublished Master Thesis, Virginia Polytechnic Institute and StateUniversity, Blacksburg, Virginia. Retrieved fromhttp://scholar.lib.vt.edu/theses/available/etd-04132001-152251/unrestricted/thesis.pdf

SIKA/SCB. (2002). Road traffic injuries 2001. Stockholm: SIKA.SNRA. (2002). All fatal accidents investigated, The SNRA’s in-depth studies – vital

data in road traffic safety work. Retrieved July 17, 2003, fromhttp://www.vv.se/traf_sak/stud_dodsolyckor/Djupstudie%204%20s%20eng.pdf

SNRA. (2003). Vision Zero. Retrieved July 17, 2003, fromhttp://www.vv.se/traf_sak/nollvis/tsnollvis3.htm

Stutts, J. C., Reinfurt, D. W., Staplin, L., & Rodgman, E. A. (2001). The role ofdriver distraction in traffic crashes. Washington DC: AAA Foundation forTraffic Safety.

Törnros, J. (1998). Driving behaviour in a real and simulated road tunnel - avalidation study. Accident & Analysis Prevention, 30(4), 497-503.

Törnros, J., Harms, L., & Alm, H. (1997). The VTI driving simulator - validationstudies (VTI-rapport No. 279, 1997). Linköping: VTI.

Wang, J.-S., Knipling, R. R., & Goodman, M. J. (1996, October, 1996). The role ofdriver inattention in crashes; new statistics from the 1995 CrashworthinessData System. Paper presented at the 40th Annual meeting of theAssociation for the advancement of automotive medicine, Vancouver.

Weber, J. W., Mullins, C. A., Schumacher, R. W., & Wright, C. D. (1994). Asystems approach to the development of an integrated crash avoidancevehicle. In 1994 Vehicle & Information Systems Conference Proceedings(pp. 431-434). New York: Institute of Electrical and Electronics Engineers.

Wickens, C. D., & Hollands, J. G. (2000). Engineering psychology and humanperformance (3rd ed.). Upper Saddle River, NJ: Prentice-Hall.

Victor, T. (2000). A technical platform for driver inattention research. RetrievedApril 17, 2003, from http://www-nrd.nhtsa.dot.gov/departments/nrd-13/driver-distraction/PDF/7.PDF

Victor, T., Blomberg, O., & Zelinsky, A. (2001). Automating driver visualbehaviour measurement. Paper presented at the Vision in Vehicles 9,Brisbane, Australia.

Wierwille, W. W. (1993). Visual and manual demands of in-car controls anddisplays. In B. Peacock & W. Karwowski (Eds.), Automotive ergonomics(pp. 299-320). London, Bristol PA: Taylor & Francis.

Wierwille, W. W., & Tijerina, L. (1996). An analysis of driving accident narrativesas a means of determining problems caused by in-vehicle visual allocationand visual workload. In A. G. Gale & al. (Eds.), Vision in Vehicles V (pp.79-86). Amsterdam: Elsevier.

Wilson, T., Butler, W., McGehee, D. V., & Dingus, T. A. (1996). IVHScountermeasures for rear-end collisions, driver warning system

REFERENCES

46

performance guidelines. In Proceedings of the 1996 Annual meeting of ITSAmerica (pp. 949-957). Washington DC: ITS America.

Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture:evidence from equiluminant visual objects. Journal of ExperimentalPsychology: Human Perception and Performance, 20(1), 95-107.

APPENDICES

47

Appendices

A – Translated written instructions

Page1

In the study you will participate, we want to know to which extentdrivers can manage other tasks while driving a car. Certainsituations may put a high demand on the driver like in urbanareas where there are a lot of other road users to consider andwhere pedestrians and cyclists often move in unpredictable ways.That type of situations does of course not leave much room forother than the driving itself. On the other hand at other times oneshould be able to carry out quite a lot of other things while driving,e.g. while driving in simpler environments such as on a freeway. Inthis study, you will only be driving on freeway.

Page 2

In the middle of the car under the air intake panel is a screenwhere arrows will show up. If there is any upward pointing arrowon the screen, press ”YES”. Otherwise press ”NO”. Do not answerbefore you are sure, because you cannot change your answer. Twoexamples with squares of arrows are shown below.

Right answer: ‘NO’

Right answer: ‘YES’

APPENDICES

48

0

Only five answers

B – Translated questionnaire with results

To be filled out by the experimenter

Date: Subject:

Questions after simulator drive

1. How many Swedish miles do you drive every year?

Swedish miles.

[Comment: A Swedish mile is 10 km]

2. a) If you drive something apart from cars, whichthen drive?

___________________________________

b) How big part of your total mileage do you drive

_____________ Swedish miles

3. For how long time have you held a driver’s license

4. How motion sick did you get driving the simulator

Not at all □ □ □ □ □

Median: 1200

Mean: 2051.7

Std: 2092.3 Max: 10,000 Min: 50 _____________

kind of vehicle do you

this kind of vehicle?

? _

?

Median: 16.5

Mean: 16.1

Std: 7.0

____________ years

□ Very

Median: 1

Mean: 1.6

Std: 1.1

APPENDICES

5. How distracted were you by the arrows task?

Not at all □ □ □ □ □ □ □ Very

6. What distracts you normally during

__________________________________

__________________________________

7. How much did you think that the aryou distracted during normal driving

Not at all □ □

8. How distracted were you when the g

Not at all □ □

Median: 3.5

Mean: 3.5

Std: 1.5

driving?

___________________________________

___________________________________

Frequency of answers mentioning:

Children: 6

Stereo/Radio/CD: 8

Cellular phone: 10

Traffic: 8

(7 subjects did not write anything)

rows task resembled things that make?

□ □ □ □ Very

Median: 4

Mean: 3.7

Std: 1.4

love department burst?

□ □ □ □ □ Very

Median: 2.5

Mean: 3.2

Std: 1.9

49

APPENDICES

50

9. How much did you think that the glove department over-flow resemblesthings that make you distracted during normal driving?

Not at all □ □ □ □ □ □ Very

10. What did you think of the strong bra

a. Freighten

Not at all □

b. Dangerou

Not at all □

c. Makes yo

Not at all □ □

d. Draws at

Not at all □ □

Median: 4

Mean: 3.8

Std: 1.7

ke pulse?

ing

□ □ □ □ □ Very

Median: 2

Mean: 2.6

Std: 1.3

s

□ □ □ □ □ Very

Median: 2

Mean: 2.2

Std: 1.4

u alert/wakes you up

□ □ □ □ Very

Median: 5

Mean: 4.8

Std: 1.3

tention upon itself

□ □ □ □ Very

Median: 5

Mean: 4.6

Std: 1.5

APPENDICES

e. Draws attention upon the road

Not at all □ □ □ □ □ □ Very

f. Easy to u

Not at all □ □

g. Warning

Not at all □ □

11. Do you think that the brake pulsdistraction system in cars?

Not at all □ □

Median: 5

Mean: 4.8

Std: 1.5

nderstand

□ □ □ □ Very

Median: 4

Mean: 3.9

Std: 1.6

□ □ □ □ Very

Median: 5

Mean: 4.5

Std: 1.7

e would be a good part of an anti-

□ □ □ □ Very

Median: 5

Mean: 4.5

Std: 1.6

51

APPENDICES

52

12. What did you think of the blue flashing light?

a. Freightening

Not at all □ □ □ □ □ □ Very

b. Dangerou

Not at all □

c. Makes yo

Not at all □ □

d. Draws at

Not at all □ □

e. Draws at

Not at all □ □

Median: 2

Mean: 2.3

Std: 1.4

s

□ □ □ □ □ Very

Median: 2

Mean: 2.0

Std: 1.2

u alert/wakes you up

□ □ □ □ Very

Median: 5

Mean: 5.0

Std: 1.7

tention upon itself

□ □ □ □ Very

Median: 6

Mean: 5.0

Std: 1.9

tention upon the road

□ □ □ □ Very

Median: 4

Mean: 4.1

Std: 1.8

APPENDICES

0

f. Easy to understand

Not at all □ □ □ □ □ □ Very

g. Warning

Not at all □ □ □ □ □ □ Very

13. Do you think that the flashing blue light would be a good part of an anti-distraction system in cars?

Not at all □ □ □ □ □ □ Very

14. How much would you be willing towhen you are distracted?

SEK ______________________

Median: 4

Mean: 4.4

Std: 1.8

Median: 4

Mean: 4.2

Std: 1.9

Median: 4

Mean: 4.6

Std: 1.7

spend on a system that warns you

Median: 1000

Mean: 2631

Std: 4312 Max: 22,500 Min:

53

APPENDICES

54

15. How did you feel about having two cameras pointed at your face?

a. Unpleasant

Not at all □ □ □ □ □ □ Very

b. You felt s

Not at all □

c. One pulls

Not at all □

16. If you summarize your impression you think the resemblance with drivi

Not at all □ □

17. How realistic did you think the roadthe road were?

Not at all □ □

Median: 1

Mean: 1.5

Std: 1

urveilled

□ □ □ □ □ Very

Median: 1

Mean: 2

Std: 1.4

oneself together because of them

□ □ □ □ □ Very

Median: 2

Mean: 2.7

Std: 1.8

of driving the simulator, how big didng a car was?

□ □ □ □ □ Very

Median: 5.5

Mean: 5.1

Std: 1.4

, the surroundings, and the objects on

□ □ □ □ Very

Median: 5

Mean: 4.5

Std: 1.4

APPENDICES

55

We have recorded your drive on video. The primary purpose is to have thepossibility to go back and watch the video if we come across anomalies in the data.But it could also be of interest for us in the future to have the possibility to use thevideo at research conferences in the future. We do not pass around the video topeople outside VTI.

Do you consent that the video is used for these purposes?

□ Yes

□ No

Date Signature

Linköping _____________ ____________________________________

APPENDICES

56

C – DiagramsC-1

C-2

Eye-tracking fallouts

0%

2%

4%

6%

8%

10%

12%

19 24 29 34 39 44

Subject

Frac

tion

of fa

llout

tim

e

Average glance duration for each S-IVIS presentation

0

0,5

1

1,5

2

2,5

3

3,5

4

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82

S-IVIS#

Tim

e (S

econ

ds)

APPENDICES

57

C-3 Combined glance time and steering angle graph

In this graph the x-axis has two functions, steering wheel angle for the full line,and glance duration time for the shaded line. The y-axis represents time. At y=2 isa dotted line representing the start of an S-IVIS presentation. At x=2 is a dottedline representing yaw-onset, and at x=3 is a dotted line representing thedistraction criterion. When the shaded line crosses the yaw onset line (i.e. whenthe subject has looked down for two seconds), the yaw starts, and when thedistraction criterion line is crossed a countermeasure sets in unless the subject isin the CONTROL condition as is the case in this particular example. Here subject#25 is subjected to its first difficult S-IVIS presentation.s

�����������������������

��������������������������

���������������������������������������������������������������������������������������������������������������

���������������������������

�������������������������

����������������������������

��������

����������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������

�������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

�������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

����������������������������������������������������������������������������������������������������������

����������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

Steering wheel angle and S-IVIS glance time

0

1

2

3

4

5

6

7

8

9

10

-12-10-8-6-4-2024681012Glance time S-IVIS/Steering angle

(seconds/degrees)

Tim

e(s

econ

ds)

LEFT RIGHT

S-IVIS presented

Distraction criterion

Turn starts

Steering angle

S-IVIS glance time

APPENDICES

58

D – Glance duration times for glove box

The glance lengths of glances on the glove box burst are given below in seconds.“Between” refers to the time between glances.

Subj Glance1 Between Glance2 Between Glance3 Between Glance4 Between Glance519 1.156 4.57 1.031 0.132 0.187 6.812 0.125 56.865 0.31320 0.718 0 0 0 0 0 0 0 021 0.781 0 0 0 0 0 0 0 022 0.875 2.065 0.563 0.264 0.594 1.67 0.625 5.317 0.12523 1.031 0.923 0.343 1.934 0.594 28.389 0.531 0 024 0.563 14.194 0.219 69.038 0.688 0 0 0 025 0.687 3.164 0.593 29.619 0.031 15.249 0.469 1.494 0.09326 0.719 83.408 0.25 0 0 0 0 0 027 0.5 1.099 0.218 34.893 0.031 28.213 0.687 15.688 1.06328 0.187 0.132 0.343 2.813 0.125 10.503 0.188 2.769 0.03129 0.812 0 0 0 0 0 0 0 030 0.781 7.295 0.093 5.098 0.469 26.543 0.125 9.229 0.65631 0.438 0.615 0.374 4.351 0.687 18.149 0.313 30.981 0.21932 0.813 1.406 1.187 10.459 2.125 11.558 1.063 7.866 1.23533 0.938 8.833 0.468 2.461 0.437 20.742 0.437 3.647 0.93834 0.469 16.479 0.375 0.352 0.5 10.063 0.281 17.007 0.81235 0.5 0 0 0 0 0 0 0 036 0.157 6.724 0.875 15.293 0.094 0.308 0.907 7.339 0.84337 0.188 0.615 0.125 66.577 0.594 0 0 0 038 0.906 38.672 0.78 40.43 0.969 3.472 0 0 039 0.406 0 0 0 0 0 0 0 040 1.188 1.055 0.312 30.762 0.062 8.218 0.375 16.128 0.03241 0.531 5.054 0.063 20.215 0.563 21.006 0.531 2.461 0.56342 1.5 4.746 0.782 7.251 0.156 2.856 0.874 1.362 0.84443 1.406 0.571 1.031 6.021 0.844 0.439 1.625 0.791 0.59444 0.281 3.779 0.219 9.976 0.531 0.703 0.188 67.544 0.18745 0.469 2.329 1.061 4.131 0.281 8.438 0.688 40.254 0.546 0.561 7.998 1.406 41.572 0.063 30.586 1.219 0 047 0.906 21.138 0.687 58.755 1.25 0.703 0.469 0.747 048 0 0 0 0 0 0 0 0 0

APPENDICES

59

All Between-values of more than 10 seconds and following glances disregarded asnot being a part of the distraction event:Subj Glance1 Between Gl.2 Bwn Gl.3 Bwn Gl.4 Bwn Gl.5 Glances19 1.156 4.57 1.031 0.132 0.187 6.812 0.125 420 0.718 121 0.781 122 0.875 2.065 0.563 0.264 0.594 1.67 0.625 5.317 523 1.031 0.923 0.343 1.934 0.594 324 0.563 125 0.687 3.164 0.593 226 0.719 127 0.5 1.099 0.218 228 0.187 0.132 0.343 2.813 0.125 329 0.812 130 0.781 7.295 0.093 5.098 0.469 331 0.438 0.615 0.374 4.351 0.687 332 0.813 1.406 1.187 233 0.938 8.833 0.468 2.461 0.437 334 0.469 135 0.5 136 0.157 6.724 0.875 237 0.188 0.615 0.125 238 0.906 139 0.406 140 1.188 1.055 0.312 241 0.531 5.054 0.063 242 1.5 4.746 0.782 7.251 0.156 2.856 0.874 1.362 543 1.406 0.571 1.031 6.021 0.844 0.439 1.625 0.791 544 0.281 3.779 0.219 9.976 0.531 0.703 0.188 445 0.469 2.329 1.061 4.131 0.281 8.438 0.688 446 0.561 7.998 1.406 247 0.906 148 0Mean 0.71 3.31 0.58 4.04 0.45 3.49 0.69 2.49 0.52 2.27

Avdelning, InstitutionDivision, Department

Institutionen för datavetenskap581 83 LINKÖPING

DatumDate2004-06-16

SpråkLanguage

RapporttypReport category

ISBN

Svenska/SwedishX Engelska/English

LicentiatavhandlingExamensarbete

ISRN LIU-KOGVET-D--04/08--SE

C-uppsats

X D-uppsatsSerietitel och serienummerTitle of series, numbering

ISSN

Övrig rapport____

URL för elektronisk versionhttp://www.ep.liu.se/exjobb/ida/2004/008/

TitelTitle

Evaluating driver distraction countermeasures

FörfattareAuthor

Rikard Karlsson

SammanfattningAbstractStatistics showing that in-vehicle driver distraction is a major contributing cause in road accidentsis presented. Driver distraction is defined building on the driving theory by Gibson and Crooks.The idea to use driver distraction countermeasures as a way of mitigating the effects of the driverdistraction problem is then introduced. A requirement list is formulated with ten requirementsthat distraction countermeasures should meet. A simplification of regarding distraction as a gazedirection problem makes way for designing an experiment to evaluate two driver distractioncountermeasures in which new eye- tracking technology plays a key role. The experiment alsomakes use of a simulator, a surrogate in-vehicle information system as a distractor, and thirtysubjects. The most important dependent measures were in-vehicle glance time and a steeringwheel reaction time measure. The evaluated countermeasures – a blue flash at middle of the roadposition and a kinesthetic brake pulse – could, however, not be shown to meet the most importantof the requirements formulated. The lack of effect of the countermeasures in the experiment mayeither depend on their actual inefficiency or on methodological shortcomings of the experiment.These alternatives are discussed. It is speculated that the biggest problems with the possible lack ofactual efficiency have to do with that the theoretical basis for using a flash did not transfer to thedriving setting, and that the brake pulse used was too weak. The methodological problems have todo with the non-validated dependent measures used, missing data, nuisance warnings,insufficient distractors, non-precise hypotheses, and difficulties with separating the effect of thecountermeasures from the psychological force to look on the road.

NyckelordKeywordsDriving Behavior, Accident Prevention, Attention, Distraction, Simulators, Eye Movements